SlideShare a Scribd company logo
1 of 120
/125
SITUATION RECOGNITION FROM
MULTIMODAL DATA
Vivek K. Singh1, Siripen Pongpaichet2, and
Ramesh Jain2
1Rutgers University,
2University of California, Irvine
/125
Today’s slides
2
http://www.springer.com/us/book/9783319305356
Or email us for a softcopy.
http://bit.ly/29JL30M
/125
Course Outline
1) Concept recognition from Multimedia data (20 mins, Ramesh Jain)
• Trends
• Why situation recognition is different from object, event, scene recognition etc.
2) Situation recognition across multiple research domains (20 mins, Vivek
Singh)
• Situation Algebra, Situation Calculus, Robotics, …
3) Situation recognition (45 mins, Vivek Singh)
• Situation modeling
• Situation operators
4) Designing situation based applications (30 mins, Siripen Pongpaichet)
• Motivation and essential requirements.
• Application scenarios: Thailand flood, hurricane Sandy, city sensing, asthma relief
5) Future trends and open problems (20 mins, Ramesh Jain)
• Future trends
• Open problems for Multimedia research
6) Question and Answers (15 mins)
3
/125
CONCEPT RECOGNITION
FROM MULTIMEDIA DATA
(20 mins, Ramesh Jain)
4
/125
Introduction
• Object, event, scene recognition
• Trends
• Situation recognition ( is different from object, event,
scene recognition etc.)
5
/125
Data, Information, Knowledge, Wisdom
6
Data is Essential.
But, we are really interested in products:
Information,
Knowledge, and
Wisdom.
/125
What is Important in ‘Big Data’?
Multimedia
Realtime Uncertainty
7
/125
The Grand Challenge
Sense making from multimodal
massive geo-social data-
streams.
8
/125
Fundamental Problem
Connecting People to Resources
effectively, efficiently, and promptly
in given situations.
/125
What is Cyber Space?
Who invented it?
Animals
Machines
Societies
Published first in 1942
10
/125
•Desired state (Goal)
•System model and Control Signal
(Actions)
•Current State (using multimedia
data)
11
/125
Input
Computed
using System
Model
Feedback
Output compared with
desired goal
Actual
System
Output
Observed
Continuously
12
/125
Social Networks
Connecting People
/125
Connecting People
And
Resources
Social Life Networks
Aggregation
and
Composition
Situation
Detection
Alerts
Queries
Information
7/21/2016 14
/125
Traditional Social Systems
•Models of Systems were difficult to
form.
•Current State of the system could
not be determined.
•Real time ‘actions’ could not be
implemented.
15
/125
•Social models can be determined
using warehouses of Big Data.
•Social observations are now
possible with little latency.
•Actions could be targeted to
precise sources.
16
/125
EventShop : Global Situation Detection
Predictive
Situation
Recognitio
n
Evolving Global Situation
Predictive
Personal
Situation
Recognitio
n
Personal EventShop
Evolving Personal Situation
Need- Resource Matcher
Recommendation
Engine
PersonaDatabas
e
Resources
Needs
Data
Ingestio
n
Wearable Sensors
Calendar
Location….
DataSources
….
Data
Ingestion
and
aggregatio
n
Database Systems
Satellite
Environmental
Sensor Devices
Social Network
Internet of Things
Actionable Information
17
/125
Concept Recognition: Last Century
18
Environ
ments
Real world
Objects
Situations
Activities
SingleMedia
SPACE
TIME
ScenesLocation
aware
Visual
Objects
Trajectories
Visual
Events
Location
unaware
Static Dynamic
Location
aware
Location
unaware
Static Dynamic
Data = Text or Images or Video
/125
Visual Concept Recognition: First research papers
• 1963: Object Recognition [Lawrence + Roberts]
• 1967: Scene Analysis [Guzman]
• 1984: Trajectory detection [Ed Chang+ Kurz]
• 1986: Event Recognition [Haynes + Jain]
• 1988: Situation Recognition [Dickmanns]
1960 1970 1980 1990 2000 2010
Object Scene
Trajectory
Event
Situation
/125
Concept Recognition: This Century
20
Environ
ments
Real world
Objects
Situations
Activities
SPACE
TIME
Location
aware
Location
unaware
Static Dynamic
HeterogeneousMedia
Location
aware
Location
unaware
Static Dynamic
Data is just Data.
Medium and sources do not matter.
/125
Concept recognition from multimedia data
21
SPACE
TIME
ScenesLocation
aware
Visual
Objects
Situations
Visual
Events
Location
unaware
Static Dynamic
HeterogeneousMedia
SingleMedia
360 K 11.4K
3.4 K
Location
aware
Location
unaware
Static Dynamic
/125
Situation Recognition: Next Frontier
• Data Abundance.
• Big Opportunity for Multimedia
• Time
• Space
• Multiple diverse data streams
• Need new frameworks.
22
/125
SITUATION RECOGNITION
ACROSS MULTIPLE
RESEARCH DOMAINS
(20 mins, Vivek Singh)
23
/125
Related Work: Data to Situations
Area Combine
hetero
data
Human
sensors
Data
analytics
Define
situations
Location
aware
Real-time
streams
Toolkits
Situation
Awareness
X X o o X
Situation
Calculus
X
Web data
mining
o X X o X
Social media
mining
o X X o X
Multimedia
Event detection
X o o o
Complex event
processing/
Active DB
X X o X
GIS X o X X o
Mashup toolkits
(Y! pipes, ifttt)
X X o X X
X
X X
X
X
X X
X
X
X
X
X
This work X X X X X X X
XX
24
o = partial support
/125
Defining Situations: Situation Calculus
• A situation s is the complete state of the universe at an
instant of time. –McCarthy, 1968
• Snapshot of the world at a given time.- Reiter, 1991
• The set of necessary and sufficient world state descriptors
for undertaking control decision. – Singh & Jain, 2009
25
/125
Situation Calculus: Quick overview
• enter(P1), startWork(P1)
• enter(P1), exit(P1), enter(P1), startWork(P1),
stopWork(P1), startWork(P1)
- isInRoom(P1, s(k))
- isWorking(P1, s(k))
isInRoom(P1, s) ˄ ~isWorking(P1, s) →
IncreaseMusicVolume()
isInRoom(P1, s) 0
isWorking(P1, s) 01
1
Situation = Not events , nor sequence of events,
but their assimilated descriptor
/125
Problems with this approach
• Scalability:
• Listing all the rules
• Frame problem - Specifying the non-effects
• Assumes 100% confidence in events detected
• Space is not a first class citizen.
• Does not deal well with heterogeneous data
27
/125
Situations
• Multiple definitions
• Situation awareness
• Situation modeling
• Situation detection
• Situation calculus
• Context based computing
“the perception of elements in the environment within a
volume of time and space, the comprehension of their
meaning, and the projection of their status in the near
future (Endsley, 1988)”.
“knowing what is going on so you can figure out
what to do” (Adam, 1993)”.
“the complete state of the universe at an instant
of time” (McCarthy, 1969)
“a set of past contexts and/or actions of individual
devices relevant to future device actions” ”
(Wang,2004)”.
“…extensive information about the environment to be
collected from all sensors independent of their interface
technology. Data is transformed into abstract symbols. A
combination of symbols leads to representation of current
situations…which can be detected”(Dietrich, 2003)
“A situation is a set of contexts in the application
over a period of time that
affects future system behavior” (Yau, 2006)
/125
Situations: commonalities
29
• Goal Based
• Space-Time
• Future Actions
• Abstraction
• Computationally
Grounded
Work Goal Based Space-Time
Future
Actions
Abstraction
Computationally
Grounded
McCarthy, 1968 X
Barwise, 1971 X X
Endsley, 1988 X X X X
Sarter, 1991 o X
Adam, 1993 X X
Dominguez,1994 X X X X
Smith, 1995 X o X X
Steinberg, 1999 X X X o
Jeannot, 2003 X
Moray, 2004 o X
Dietrich, 2004 X X
Yau, 2006 X X X
Dostal, 2007 o X
Singh, 2009 X X X
Merriam-Webster
(accessed 2012) o
This work (aim) X X X X X
o = Partial support
/125
Situation: Definition
• Situation: An actionable abstraction of
observed spatio-temporal characteristics.
• e.g. flu epidemic, severe asthma threat, road
congestion, wildfire, flash-mob
30
Goal Based Space-Time
Future
Actions
Abstraction
Computationally
Grounded
/125
Overall Framework: Motivating example
31
STT data
Tweet:
‘Urrgh… sinus’
Loc: NYC,
Date: 3rd Jun, 2011
Theme: Allergy
Situation Detection User-Feedback
‘Please visit nearest CDC
center at 4th St
immediately’
Date: 3rd Jun, 2011
Aggregation,
1) Classification
2) Control action
Operations
Alert level
= High
/125
Applications
• Healthcare
• Alert me if there is a flu epidemic in my area
• Telepresence:
• Which camera feed to send out?
• Business analysis:
• Where is the most suitable place to open a new ‘iphone’ store ?
• Weather
• Alert me when the fall colors blossom in New England?
• Daily living:
• Which place (and at what time) is conducive for exercising?
• Weather, climate, politics, traffic, …
32
/125
SITUATION RECOGNITION
FRAMEWORK
(45 mins, Vivek Singh)
33
/125
A) Situation
Modeling
B) Situation
Recognition
C) Visualization,
Personalization, and Alerts
…
STT
Stream
Emage
Situation
C1

v2 v3

v5 v6
@
∏
Δ
@
i) Visualization
ii) Personalization
+
+
Available
resources
iii) Alerts
Personal
context
Personal
ized
situation
Overall framework 34
/125
A) Situation Modeling
• Help domain experts externalize their internal
models of situations of interest e.g. epidemic.
• Building blocks:
• Operators
• Operands
• Wizard:
• A prescriptive approach for modeling situations using
the operators and operands
35
Singh, Gao, Jain: Situation recognition: An evolving problem for heterogeneous
dynamic big multimedia data, ACM Multimedia ‘12.
/125
Growth rate
(Flu reports)
Feature
Thresholds
(0, 50)
Data source
Meta-data
-Emage
(#Reports)
Representation
level
Twitter-Flu
Building Blocks: Operands
36
• Knowledge or data driven building blocks
/125
Building Blocks: Operators
Δ Transform …
Spatio-temporal
window
37
 Aggregate +
 Classification
Classification
method
@ Characterization Growth Rate
= 125%
Property
required
Pattern Matching
72%
+
Pattern
∏ Select +
Mask
Φ Learn Learning
method
{Features}
{Situation}
f f
1) Data into right
representation
2) Analyze data to
derive features
3) Use features to
evaluate situations
Supporting
parameter(s)
Data OutputOperator Type
/125
Situation Modeling
Get_components (Situation v){
1) Identify output state space
2) Identify S-T bounds
3) Define component features:
v=f(v1, …, vk)
• If (type = imprecise)
• identify learning data source, method
4) ForEach (feature vi) {
If (atomic)
• Identify Data source.
• Type, URL, ST bounds
• Identify highest Rep. level reqd.
• Identify operations
Else
Get_components(vi)
}
}
38
v
f1
v4

v2 v3
@
D1
Emage
Δ
D2
∏
Emage
Δ
D3
Δ
@
Emage
D2
∏
Emage
Δ
f2 
v5 v6
<USA, 5 mins,
0.01x 0.01>
ϵ { Low,
Mid, High}
/125
Epidemic
Outbreaks
Unusual
Activity? Growth Rate

Current activity
level
Historical
activity level

Emage
(#reports ILI)
Δ
Twitter-Flu

Twitter.com
<USA, 5 mins,
0.01x 0.01>
Emage
(Historical avg)
Δ
Twitter-Avg
DB,
<USA, 5 mins,
0.01x 0.01>
Δ
Twitter-Flu
Emage
(#reports ILI)
Twitter.com
<USA, 5 mins,
0.01x 0.01>
ϵ {Low, mid, high},
<USA, 5 mins, 0.01x
0.01>
Growing Unusual
activity

1)Model
Emage
(#reports ILI)
Δ
Twitter-Flu
Emage
(population)
Δ
CSV-
Population

π
Twitter.com
<USA, 5 mins,
0.01x 0.01>
Census.gov,
<USA, 5 mins,
0.01x 0.01>
2) Revise
Subtract
Subtract
Multiply
Classification:
Thresh (30,70)
Normalize
[0,100]
3) Instantiate
39
/125
Level 1: Unified
representation
(STT Data)
Level 3:
Symbolic rep.
(Situations)
Properties
Properties
Properties
Level 0: Raw data streams
e.g. tweets, cameras, traffic, weather, …
Level 2:
Aggregation
(Emage)
…
STT Stream
Emage
Situation
B) Situation evaluation: Workflow
40
Operations
/125
Data Representation
• E-mage
• Visualization
• Spatio temporal data representation
• Data analysis using media processing operators
(e.g. segmentation, background subtraction,
convolution)
41
/125
Data Model
• Spatio-temporal element
• stel = [s-t-coord, theme(s), value(s), pointer(s)]
• E-mage
• g = (x, {(tm, v(x))}|xϵ X = R2 , tm ϵ θ, and v(x) ϵ V = N)
• Temporal E-mage Set
• TES= {(t1, g1), ..., (tn, gn)},
• Temporal Pixel Set
• TPS = {(t1, p1), ..., (tn, pn)},
/125
Situation Recognition Algebra
Singh, Gao, Jain: Social Pixels: Genesis and Evaluation, ACM Multimedia ‘10.
43
S. No Operator Input Output
1 Filter ∏ Temporal E-mage Stream Temporal E-mage Stream
2 Aggregation  K*Temporal E-mage Stream Temporal E-mage Stream
3 Classification  Temporal E-mage Stream Temporal E-mage Stream
4 Characterization : @
 Spatial
 Temporal
 Temporal E-mage Stream
 Temporal Pixel Stream
 Temporal Pixel Stream
 Temporal Pixel Stream
5 Pattern Matching 
 Spatial
 Temporal
 Temporal E-mage Stream
 Temporal Pixel Stream
 Temporal Pixel Stream
 Temporal Pixel Stream
/125
Media
processing
engine
/125
Implementation and results
• Twitter feeds
• Geo-coding user home location
• Loops of location based queries for different terms
• Over 100 million tweets using ‘Spritzer’ stream (since Jun 2009),
and the higher rate ‘Gardenhose’ stream since Nov, 2009.
• Flickr feeds
• API
• Tags, RGB values from >800K images
/125
Testing Data Representation + Algebra
• Applications
• Business analytics
• Political event analytics
• Seasonal characteristics
• Data
• Twitter feeds archive
• Loops of location based queries for different terms
• Over 100 million tweets using ‘Spritzer’/ ‘Gardenhose’ APIs
• Flickr feeds
• API: Tags, RGB values from >800K images
• Implementation
• Matlab + Java + Python
46
/125
Sample Queries
• Select E-mages of USA for theme ‘Obama’.
• ∏spatial(region=[24,-125],[24,-65]) (TEStheme=Obama)
• Identify three clusters for each E-mage above.
• kmeans(3) (∏spatial(region=[24,-125],[24,-65])(TEStheme=Obama))
• Show me the cluster with most interest in ‘Obama’.
• ∏value(v=1) (kmeans(n=3) (∏spatial(region=[24,-125],[24,-65]) (TEStheme=Obama)))
• Show me the speed for high interest cluster in ‘Katrina’ emages
• @speed(@epicenter(∏value(v=1) (kmeans(n=3) (∏spatial(region=[24,-125],[24,-65])
(TEStheme=Katrina)))))
• How similar is pattern above to ‘exponential increase’?
• exp-increase(@speed(@epicenter (∏value(v=1) (kmeans(n=3) (∏spatial(region=[24,-125],[24,-65])
(TEStheme=Katrina))))
47
/12548
AT&T
retail
locations
AT&T total
catchment
area
iPhone theme
based e-mage,
Jun 2
Aggregate
interest
Under-served
interest areas
-
Subtract
Decision
Best Location is at
Geocode [39, -
122] , just north of
Bay Area, CA
@Spatial.Max
<geoname>
<name>College City</name>
<lat>39.0057303</lat>
<lng>-122.0094129</lng>
<geonameId>5338600</geonameId>
<countryCode>US</countryCode>
<countryName>United
States</countryName>
<fcl>P</fcl>
<fcode>PPL</fcode>
<fclName>city, village,...</fclName>
<fcodeName>populated
place</fcodeName>
<population/>
<distance>1.0332</distance>
</geoname>
+ Add
to Jun 15, 2009
Convolution
.
*
Store
catchment
area
Convolution.
*
Store catchment
area
/125
Flickr Social Emages
• Jan – Dec 2009
/125
Seasonal characteristics analysis
• Fall colors in New England
• Show me the difference between red and green colors for New
England region, as it varies throughout the year.
• subtract(@spatial(sum)(πspatial(R=[(40,-76), (44,-71)]) (TEStheme=Red)),
@spatial(sum)(πspatial(R=[(40,-76), (44,-71)])(TEStheme=Green)))
50
Jan
0
Dec
/125
Year average Peak of green
At [35, -84], at the junction of Chattahoochee National Forest, Nantahala
National Forest, Cherokee National Forest and Great Smoky
Mountains National Park
/125
OTHER APPROACHES
52
/125
FraPPE: a vocabulary to represent heterogeneous
spatio-temporal data to support visual analytics
Marco Balduini, Emanuele Della Valle
ISWC 2015 – Data Sets and Ontologies
Slides adapted from:
http://www.slideshare.net/MarcoBalduini/frappe-a-vocabulary-to-represent-heterogeneous-spatiotemporal-data-to-support-visual-analytics
/125
Proposed Approach
54
21/07/2016
Re-use consolidated concepts from
• geo-spatial vocabularies
• time related vocabularies
• provenance vocabularies
Model visual analytics concepts *
• pixels
• frame
Fill the gap between heterogeneous geo-spatial time-varying data and visual
analytics concepts to create actionable information and ease the decision
making processes of final users
* Singh, V.K., Gao, M., Jain, R.: Social pixels: genesis and evaluation (ACM MM 2010)
/125
FraPPE, visually
55
21/07/2016
Reality
Capture
Frame
/125
FraPPE, visually
56
21/07/2016
Grid
Cell
Frame
/125
FraPPE, visually
57
21/07/2016
Pixel Frame 1
/125
FraPPE, visually
58
21/07/2016
Place A
Event A
/125
FraPPE, visually
59
21/07/2016
Event A
Frame 1
/125
FraPPE, visually
60
21/07/2016
Event B
Place B
Frame 2
Frame 1
/125
City Sensing listens to the pulse of Milano
Design Week on April 9th, 2014
61
/125
Tweeting Cameras
62
Slides courtesy:
Yuhui Wang, Francesco Gelli, and Mohan Kankanhalli
Adapted from “Tweeting Cameras for
Event Detection” in Proc. WWW
Conference 2013.
/125
Physical & Social Sensors Fusion For
Situation Awareness
Physical
Sensors
Social
Sensors
/125
Real-world Events
Hispanic ParadeCBGB Musical Festival
Columbus Day Parade
twtw
tw tw
tw tw
twtw
tw tw
tw tw
twtw
tw tw
tw tw
twtw
tw tw
tw tw
Historic TweetsRecent Tweets
Event time and location
Retrieve recent tweets Retrieve historic tweets
/125
Probabilistic Spatio-Temporal Data
• Definition (PST: Probabilistic Spatio-Temporal Data) The
fundamental building block for low-level concept
representation is the probabilistic spatio-temporal element
“pst”.
• pst = [loc, temp, label, prob, pointer] (1)
where
• loc = [lat, lon] represents the geo-location – latitude and longitude – of
the camera location. • temp stores the time information of captured data.
• label represents semantic concept such as car, human,crowd, parade,
etc., detected in the stream. Generally, these concepts express low-level
abstraction of information which could be semi-reliably detected by existing
detectors or classifiers.
• prob is the confidence value in [0,1] representing the output of a concept
detector as a probability value.
• pointer points to actual raw data stream.
65
/125
Physical Sensors (Concept: “Crowd”)
/125
Social Sensors (#MillionMarchNYC, #BlackLivesMatter)
/125
Fused Information
/125
CMage
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
Gaussian Process
based Prediction
Sensor Image Patch
/125
Fusing Sensor Cmage with Social Cmage
Sensor Cmage
(concept: “people marching”)
Social Cmage
(concept: “MillionsMarchNYC”)
Fused Cmage
/125
DESIGNING SITUATION
BASED APPLICATIONS
(30 mins, Siripen Pongpaichet)
71
/125
Outline
• EventShop System Requirements
• EventShop System Architecture
• Demo
• Building Applications using EventShop
• Conclusion
72
/125
EventShop Requirement
73
Granulari-
ties
Heterogen
eous
Model Prediction
Users
Open-
Source
Storage
Generic
Streams
Support fast data flow
Handle heterogeneous
types of data streams
Efficiently aggregate data at
different granularities
Provide storage system to
archive both data input and
system output.
Create situation model and
provide actionable information
Generic computational platform
for situation recognition
Open-source software
User friendly and interactive
interface
Contain predictive component
/125
EventShop Architecture
74
Alert/OutputData Ingestor
Data Source
Parser
Data Adapter
Emage
Generator
(+resolution mapper)
Query Processing
EventShop Storage
Query
Parser
Query
Rewriter
Emage Stream Processing
Action Parser
Situation
Emage
Visualization
(Dashboard)
Actuator
Communication
Event Property &
Other Information
(e.g., spatio-temporal
pattern)
ᴨ
ᴨ
µ
Data Access Manager
Live Stream
Archived Stream
Situation Stream
Physical
Data Source
(e.g., sensor
streams, geo-image
streams)
Logical Data Source
(e.g., preprocessing
data streams, social
media streams)
Raw Event
REST API Services
Data Source / Query / Alerts / STT-Emage
EventShop UI External AppsVisual Analytics
InterfaceAPIProcessingLayerStorageLayer
/125
EventShop UI
75
Save Query Reset Query
Create Query
Pollen
Tweets_
Asthma
Available
Available
350
AQI Available357
361
Grouping Stopped35
Asthma_
Risk
Stopped36
Asthma_
Interpolate
Stopped37
Asthma_
Interpolate
Stopped37
Asthma_ Stopped37
Query Graph
Filter Group Aggregation Spatial Pattern Temporal Pattern Spatial Char Temporal Char
Redraw
ds361
ds350
ds357
F1 Q1
F2
F3
Q2
Q3
A1 Q4 G1 Q5
http://eventshop.ics.uci.edu:8080/eventshoplinux/
/125
Demo
76
https://www.youtube.com/watch?v=E5unHXZmSr8
/125
Demo
77
https://www.youtube.com/watch?v=IwJYEZd8Bbg
/125
Building applications using EventShop
S.No Application Data Used
Application
deployed?
Scale Data modalities Operators used
1
Wildfire detection in
California
Real Yes Macro
Satellite data,
Google insights
F, A, Ch
2 Hurricane monitoring Simulated No Macro n/a F, A, Ch, P
3
Flu epidemic
surveillance
Real No Macro Twitter, Census F, A, C
4 Allergy recommendation Real Yes Macro
Twitter, Air Quality,
Pollen Count
F, A, C
5 Asthma management Real Yes
Macro,
Personalized alerts
In situ sensors,
Satellite data,
Asthma Tracking
F, I, Pr
6. Thailand flood mitigation Real Yes
Macro,
Personalized alerts
KML F, A, C
7.
Photos as Micro-
Reports
Real Yes Macro Flickr F, Cl
8. Trash management
Real &
Simulated
In progress Macro
Trash sensors,
micro-reports
F, A, Pr,
78
/125
Asthma
Management
Application
79
/125
Asthma Management Application
80
(1)
Macro
Situation
Macro
Data Streams
(3)
Situation-
Action Rules
Sensor streams
Social media
Geo-temporal data
Personal
Data Streams
(2)
Personal
Situation
Behavioral
streams
Profile +
Preferences
/125
Asthma Risk Estimation
81
Traffic Flow Aerosol Concentration PM2.5 CMAQ Model PM2.5 Concentration
*visualize data on Feb 12th, 2008
Mengfan Tang, Pranav Agrawal, Siripen Pongpaichet, Ramesh Jain:
Geospatial interpolation analytics for data streams in EventShop. ICME 2015
Spectral Spatial Gaussian Process (SSGP)
/125
Experiment Results
82
Data Model PMSE MAPE
Single
Data Source
CMAQ - 1.0619 27.2873
CMAQ LR 0.9586 27.1077
Stations Kriging 0.9077 22.9672
CMAQ SSGP 0.3468 14.2727
Multiple
Data Sources
ALL SGP 0.3006 13.5109
ALL SSGP 0.2858 13.1087
CMAQ Kriging SSGP
/125
Asthma Risk Estimator Model and Result
83
Asthma Hospitalization
Ground Truth
FILTER
LOC=CA
FILTER
LOC=CA
AGG
FUNC=AVG
GROUP
THRESHOLD
Asthma Risk Area
without Interpolation
GROUP
THRESHOLD
Asthma Risk Area
with interpolation
AGG
FUNC=AVG
PM2.5
Concentration
From Stations
Interpolated
PM2.5
using SSGP
Pollen
Ozone
AQI
/125
A GRAPH BASED MULTIMODAL
GEOPATIAL INTERPOLATION FRAMEWORK
Mengfan Tang1, Pranav Agrawal1, Feiping Nie2,
Siripen Pongpaichet1, Ramesh Jain1
1University of California, Irvine, USA
2Northwestern Polytechnical University, China
Tuesday July 12th, 2016 at 5PM Room: Cascade I,
Special Session: Multimedia Cloud Computing and Big Data
/125
Detecting
Situations from
Micro-Reports
85
Photos
Reports Events
/125
PHOTOS as Kodak Moments
/125
Disruption: PHOTOS as Information
Smartphone camera
captures
EVENTS
/125
• Truthfulness,
• Accuracy,
• Objectivity,
• Fairness and Public accountability
Reports of Events from Journalists
Seek Truth and Report it as Fully as Possible
/125
Reports of Events from Citizens
/125
FASTSubjective
EASY Noisy
LATES
T
Ambiguous
were SO yesterday!Micro-Blogs
Multimedia
Micro-Reports
/125
Compelling Universal
Objective Spontaneous
Multimedia Micro-Reports (MMRs)
are now and future
/125
Capturing and Reporting events using multimedia
such as photos, videos, sensors, and texts
Converting multimedia data to multimedia
micro-reports using MediaJSON
Integrating multimedia micro-reports with
other data sources for situation recognition,
trend analysis, and culture analytics.
Emerging opportunities for numerous apps:
smart city, public health, emergency rescue
What are the challenges?
/125
Capturing and Reporting Events
with Krumbs SDK
https://krumbs.net/
What: Objects
Who: People
When: Events
Where: Location
Why: Intent/Emotions
How: Photo and audio
/125
Capturing and Reporting Events
with Krumbs SDK
https://krumbs.net/
/125
Real-time MMR Dashboard
/125
Converting Multimedia Data into MMR
{"micro_reports":[{
"where":{
"geo_location":{
"latitude":32.90233332316081,
"longitude":-117.2441166718801},
"when":{
"start_time":"Jun 14, 2009 11:25:19
AM",
"end_time":"Jun 14, 2009 11:25:19 AM",
"time_zone":"America/Los_Angeles"},
"what":[{
"concept_name":"people",
"confidence":0.9836078882217407,
"visual_concept_provider":"CLARIFAI"},
… {
"concept_name":"food",
"confidence":0.8526291847229004,
"visual_concept_provider":"CLARIFAI"}]
,
"tag":”#niceday #summer",
"source":{"default_src":"https://….jpg"}},
"sub_event":[],
"why":[]},
…]}
Photo
What
Where
When
Who
Why
Sound
MediaJSON
/125
MediaJSON
Data
Wrapper
Data
Wrapper
Data
Wrapper
Data
Wrapper
IoT
(Event-driven operation)
Converting Multimedia Data into MMR
MediaJSON
Data
Wrapper
/125
Number of photos in London per day
/125
Evolving Photo Concepts in London
/125
• Year of interest: year 2008 and 2012
• Training locations: Beijing (China)
• Testing location: London (UK)
• Create “Olympic Games” Event Model from
“BAG of Visual Concepts”
Detecting Olympic Games
Olympic Games = {basketball, court game,
gymnastics, people, sport, stadium, swim, tennis}
/125
Evolving Photo Concepts in Beijing
Event model of “Olympic Game”
/125
Detecting London Olympic Games
Summer Olympic Game
in July and August
Paralympic
Games in
September
/125
• Temporal range:
1 year from July 2011 to
June 2012
• Location: Thailand
Detecting Emergency Situations
City flood = {outdoor, water, road, car}
/125
Photos from City Flood Cluster
November 5, 2011
/125
Smart City
Project in DC
105
U.S. Presidential Inauguration in
DC
Earth Day Concert
Cherry Blossom Festival
/125
Integrating MMR with other data sources for
Situation Recognition (In progress)
h t t p : / / s m a r t c i t i e s i n n o v a t i o n . c o m /
/125
Trash Fill Level Situation in DC
107
/125
0
35
50
90
0
20
40
60
80
100
7:30 8:00 8:30 9:00 9:30 10:00 10:30
Real-Time Fill Level Situations at a given location of an event
Prediction based on Events History
Events Data
Real-Time Trash Fill Level Situation
20 42
Now
Predicted Trash Fill Level
in 30 minutes at a given location
78 99
30 minutes
10
30
40
70
90
100
20
0
20
40
60
80
100
120
7:30 8:00 8:30 9:00 9:30 10:00 10:30
Projected Trash Fill Level at a given location based on Event
History
/125
Conclusion
• EventShop Architecture
• Situation-based Applications using EventShop
• Online Service: http://sln.ics.uci.edu:8085/eventshoplinux
• Open Source: http://dabuntu.github.io/es/
• For more information about EventShop including tutorial
videos, presentations and publications, please visit my
home page http://www.ics.uci.edu/~spongpai
Siripen Pongpaichet (spongpai@uci.edu)
109
/125
FUTURE TRENDS AND
OPEN PROBLEMS
(20 mins, Ramesh Jain)
110
/125
Future Trends
• Future trends
• Open problems for Multimedia research
111
/125
This century is different from the last.
Should we think differently???
/125In 20th century, we tolerated photos
in our textual documents.
In 21st century, you create visual documents
that tolerate text.
/125
Major Disruption in Photos: From Memories
to Information Sources.
Photos are the most compelling source of
information.
/125115
/125
We are immersed in Big Data.
Multimedia
Realtime Uncertainty
116
/125117
/125
Data as a Platform.
• Multi-modal
• Multimedia has to become multimodal
• Data Streams
• Important things – Situation recognition
• Real time action for
118
Connecting People to Resources
effectively, efficiently, and promptly
in given situations.
/125
Contact Information
• Vivek Singh, Rutgers University
• Vivek.k.singh@Rutgers.edu
•
• Siripen Pongpaichet
• spongpai@ics.uci.edu
• Ramesh Jain
• jain@ics.uci.edu
119
/125
Useful links
• Copies of publications
• http://wp.comminfo.rutgers.edu/vsingh/publication/
• Today’s Slides:
• https://dl.dropboxusercontent.com/u/5887580/Tutorial_SituationRecog.
pdf
• EventShop
• Online Service: http://sln.ics.uci.edu:8085/eventshoplinux
• Open Source: http://dabuntu.github.io/es/
• For more information about EventShop including tutorial videos,
presentations and publications: http://www.ics.uci.edu/~spongpai
• Related Projects
• Tweeting Cameras: https://sites.google.com/site/fredyuhuiwang/
• Frappe:
http://www.streamreasoning.org/live/festivalcomunicazione2014/
120

More Related Content

What's hot

Hennig_AgileProject_1PageOverview
Hennig_AgileProject_1PageOverviewHennig_AgileProject_1PageOverview
Hennig_AgileProject_1PageOverview
Anthony Hennig
 
Use Machine Learning to Get the Most out of Your Big Data Clusters
Use Machine Learning to Get the Most out of Your Big Data ClustersUse Machine Learning to Get the Most out of Your Big Data Clusters
Use Machine Learning to Get the Most out of Your Big Data Clusters
Databricks
 
Data Data Everywhere: Not An Insight to Take Action Upon
Data Data Everywhere: Not An Insight to Take Action UponData Data Everywhere: Not An Insight to Take Action Upon
Data Data Everywhere: Not An Insight to Take Action Upon
Arun Kejariwal
 

What's hot (10)

NASA Data Science Day Plenary: Applied Machine Learning (ML)
NASA Data Science Day Plenary: Applied Machine Learning (ML)NASA Data Science Day Plenary: Applied Machine Learning (ML)
NASA Data Science Day Plenary: Applied Machine Learning (ML)
 
Multipleregression covidmobility and Covid-19 policy recommendation
Multipleregression covidmobility and Covid-19 policy recommendationMultipleregression covidmobility and Covid-19 policy recommendation
Multipleregression covidmobility and Covid-19 policy recommendation
 
10 Steps to Optimize Your Crime Analysis
10 Steps to Optimize Your Crime Analysis10 Steps to Optimize Your Crime Analysis
10 Steps to Optimize Your Crime Analysis
 
Hennig_AgileProject_1PageOverview
Hennig_AgileProject_1PageOverviewHennig_AgileProject_1PageOverview
Hennig_AgileProject_1PageOverview
 
Emotion Sense: From Design to Deployment
Emotion Sense: From Design to DeploymentEmotion Sense: From Design to Deployment
Emotion Sense: From Design to Deployment
 
Examples of Applied Semantic Technologies: Application of Semantic Sensor Net...
Examples of Applied Semantic Technologies: Application of Semantic Sensor Net...Examples of Applied Semantic Technologies: Application of Semantic Sensor Net...
Examples of Applied Semantic Technologies: Application of Semantic Sensor Net...
 
Artificial intelligence in space exploration venkat vajradhar - medium
Artificial intelligence in space exploration   venkat vajradhar - mediumArtificial intelligence in space exploration   venkat vajradhar - medium
Artificial intelligence in space exploration venkat vajradhar - medium
 
Use Machine Learning to Get the Most out of Your Big Data Clusters
Use Machine Learning to Get the Most out of Your Big Data ClustersUse Machine Learning to Get the Most out of Your Big Data Clusters
Use Machine Learning to Get the Most out of Your Big Data Clusters
 
Big Data to avoid weather related flight delays
Big Data to avoid weather related flight delaysBig Data to avoid weather related flight delays
Big Data to avoid weather related flight delays
 
Data Data Everywhere: Not An Insight to Take Action Upon
Data Data Everywhere: Not An Insight to Take Action UponData Data Everywhere: Not An Insight to Take Action Upon
Data Data Everywhere: Not An Insight to Take Action Upon
 

Similar to Situation Recognition from Multimodal Data Tutorial (ICME2016)

WWW2010_Earthquake Shakes Twitter User: Analyzing Tweets for Real-Time Event...
WWW2010_Earthquake Shakes Twitter User: Analyzing Tweets for Real-Time Event...WWW2010_Earthquake Shakes Twitter User: Analyzing Tweets for Real-Time Event...
WWW2010_Earthquake Shakes Twitter User: Analyzing Tweets for Real-Time Event...
tksakaki
 
Gis120 lec1 slide_share_practice
Gis120 lec1 slide_share_practiceGis120 lec1 slide_share_practice
Gis120 lec1 slide_share_practice
Michelle Kinzel
 

Similar to Situation Recognition from Multimodal Data Tutorial (ICME2016) (20)

Building Social Life Networks 130818
Building Social Life Networks 130818Building Social Life Networks 130818
Building Social Life Networks 130818
 
Scenario Methodology for Planning Future Activities
Scenario Methodology for Planning Future ActivitiesScenario Methodology for Planning Future Activities
Scenario Methodology for Planning Future Activities
 
Bridging the Gap: Machine Learning for Ubiquitous Computing -- ML and Ubicomp...
Bridging the Gap: Machine Learning for Ubiquitous Computing -- ML and Ubicomp...Bridging the Gap: Machine Learning for Ubiquitous Computing -- ML and Ubicomp...
Bridging the Gap: Machine Learning for Ubiquitous Computing -- ML and Ubicomp...
 
Multimedia rescue 161018
Multimedia rescue 161018Multimedia rescue 161018
Multimedia rescue 161018
 
Scenarios for Risk and Disaster Management
Scenarios for Risk and Disaster ManagementScenarios for Risk and Disaster Management
Scenarios for Risk and Disaster Management
 
Spatio Temporal Data Mining
Spatio Temporal Data MiningSpatio Temporal Data Mining
Spatio Temporal Data Mining
 
WWW2010_Earthquake Shakes Twitter User: Analyzing Tweets for Real-Time Event...
WWW2010_Earthquake Shakes Twitter User: Analyzing Tweets for Real-Time Event...WWW2010_Earthquake Shakes Twitter User: Analyzing Tweets for Real-Time Event...
WWW2010_Earthquake Shakes Twitter User: Analyzing Tweets for Real-Time Event...
 
Micro reports and Situation Recognition at social machines workshop
Micro reports and Situation Recognition at social machines workshopMicro reports and Situation Recognition at social machines workshop
Micro reports and Situation Recognition at social machines workshop
 
Christoph Barrett - Policy Informatics at Societal Scale
Christoph Barrett - Policy Informatics at Societal ScaleChristoph Barrett - Policy Informatics at Societal Scale
Christoph Barrett - Policy Informatics at Societal Scale
 
4th Workshop on Strategic Crisis Management, Presentation, Panel 3 - anticipa...
4th Workshop on Strategic Crisis Management, Presentation, Panel 3 - anticipa...4th Workshop on Strategic Crisis Management, Presentation, Panel 3 - anticipa...
4th Workshop on Strategic Crisis Management, Presentation, Panel 3 - anticipa...
 
The Consequences of Living and Breathing with Hyperconnectedness
The Consequences of Living and Breathing with HyperconnectednessThe Consequences of Living and Breathing with Hyperconnectedness
The Consequences of Living and Breathing with Hyperconnectedness
 
Applications of AI in the geospatial domain
Applications of AI in the geospatial domainApplications of AI in the geospatial domain
Applications of AI in the geospatial domain
 
Context is Highly Contextual
Context is Highly ContextualContext is Highly Contextual
Context is Highly Contextual
 
OEM Presentation - IA and Emergency Response
OEM Presentation -  IA and Emergency ResponseOEM Presentation -  IA and Emergency Response
OEM Presentation - IA and Emergency Response
 
context aware.pptx
context aware.pptxcontext aware.pptx
context aware.pptx
 
Ethics and technology in humanitarian setting
Ethics and technology in humanitarian settingEthics and technology in humanitarian setting
Ethics and technology in humanitarian setting
 
Contextual Dissonance: Design Bias in Sensor-Based Experience Sampling Methods
Contextual Dissonance: Design Bias in Sensor-Based Experience Sampling MethodsContextual Dissonance: Design Bias in Sensor-Based Experience Sampling Methods
Contextual Dissonance: Design Bias in Sensor-Based Experience Sampling Methods
 
Making sense
Making senseMaking sense
Making sense
 
Situation based analysis and control for supporting Event-web applications
Situation based analysis and control for supporting Event-web applicationsSituation based analysis and control for supporting Event-web applications
Situation based analysis and control for supporting Event-web applications
 
Gis120 lec1 slide_share_practice
Gis120 lec1 slide_share_practiceGis120 lec1 slide_share_practice
Gis120 lec1 slide_share_practice
 

Recently uploaded

Jual Obat Aborsi Hongkong ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...
Jual Obat Aborsi Hongkong ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...Jual Obat Aborsi Hongkong ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...
Jual Obat Aborsi Hongkong ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...
ZurliaSoop
 
Vishram Singh - Textbook of Anatomy Upper Limb and Thorax.. Volume 1 (1).pdf
Vishram Singh - Textbook of Anatomy  Upper Limb and Thorax.. Volume 1 (1).pdfVishram Singh - Textbook of Anatomy  Upper Limb and Thorax.. Volume 1 (1).pdf
Vishram Singh - Textbook of Anatomy Upper Limb and Thorax.. Volume 1 (1).pdf
ssuserdda66b
 
Activity 01 - Artificial Culture (1).pdf
Activity 01 - Artificial Culture (1).pdfActivity 01 - Artificial Culture (1).pdf
Activity 01 - Artificial Culture (1).pdf
ciinovamais
 
The basics of sentences session 3pptx.pptx
The basics of sentences session 3pptx.pptxThe basics of sentences session 3pptx.pptx
The basics of sentences session 3pptx.pptx
heathfieldcps1
 

Recently uploaded (20)

ICT Role in 21st Century Education & its Challenges.pptx
ICT Role in 21st Century Education & its Challenges.pptxICT Role in 21st Century Education & its Challenges.pptx
ICT Role in 21st Century Education & its Challenges.pptx
 
How to Give a Domain for a Field in Odoo 17
How to Give a Domain for a Field in Odoo 17How to Give a Domain for a Field in Odoo 17
How to Give a Domain for a Field in Odoo 17
 
On National Teacher Day, meet the 2024-25 Kenan Fellows
On National Teacher Day, meet the 2024-25 Kenan FellowsOn National Teacher Day, meet the 2024-25 Kenan Fellows
On National Teacher Day, meet the 2024-25 Kenan Fellows
 
HMCS Max Bernays Pre-Deployment Brief (May 2024).pptx
HMCS Max Bernays Pre-Deployment Brief (May 2024).pptxHMCS Max Bernays Pre-Deployment Brief (May 2024).pptx
HMCS Max Bernays Pre-Deployment Brief (May 2024).pptx
 
Explore beautiful and ugly buildings. Mathematics helps us create beautiful d...
Explore beautiful and ugly buildings. Mathematics helps us create beautiful d...Explore beautiful and ugly buildings. Mathematics helps us create beautiful d...
Explore beautiful and ugly buildings. Mathematics helps us create beautiful d...
 
Graduate Outcomes Presentation Slides - English
Graduate Outcomes Presentation Slides - EnglishGraduate Outcomes Presentation Slides - English
Graduate Outcomes Presentation Slides - English
 
Jual Obat Aborsi Hongkong ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...
Jual Obat Aborsi Hongkong ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...Jual Obat Aborsi Hongkong ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...
Jual Obat Aborsi Hongkong ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...
 
Vishram Singh - Textbook of Anatomy Upper Limb and Thorax.. Volume 1 (1).pdf
Vishram Singh - Textbook of Anatomy  Upper Limb and Thorax.. Volume 1 (1).pdfVishram Singh - Textbook of Anatomy  Upper Limb and Thorax.. Volume 1 (1).pdf
Vishram Singh - Textbook of Anatomy Upper Limb and Thorax.. Volume 1 (1).pdf
 
Activity 01 - Artificial Culture (1).pdf
Activity 01 - Artificial Culture (1).pdfActivity 01 - Artificial Culture (1).pdf
Activity 01 - Artificial Culture (1).pdf
 
Google Gemini An AI Revolution in Education.pptx
Google Gemini An AI Revolution in Education.pptxGoogle Gemini An AI Revolution in Education.pptx
Google Gemini An AI Revolution in Education.pptx
 
General Principles of Intellectual Property: Concepts of Intellectual Proper...
General Principles of Intellectual Property: Concepts of Intellectual  Proper...General Principles of Intellectual Property: Concepts of Intellectual  Proper...
General Principles of Intellectual Property: Concepts of Intellectual Proper...
 
How to Create and Manage Wizard in Odoo 17
How to Create and Manage Wizard in Odoo 17How to Create and Manage Wizard in Odoo 17
How to Create and Manage Wizard in Odoo 17
 
The basics of sentences session 3pptx.pptx
The basics of sentences session 3pptx.pptxThe basics of sentences session 3pptx.pptx
The basics of sentences session 3pptx.pptx
 
Making communications land - Are they received and understood as intended? we...
Making communications land - Are they received and understood as intended? we...Making communications land - Are they received and understood as intended? we...
Making communications land - Are they received and understood as intended? we...
 
Unit-IV- Pharma. Marketing Channels.pptx
Unit-IV- Pharma. Marketing Channels.pptxUnit-IV- Pharma. Marketing Channels.pptx
Unit-IV- Pharma. Marketing Channels.pptx
 
FSB Advising Checklist - Orientation 2024
FSB Advising Checklist - Orientation 2024FSB Advising Checklist - Orientation 2024
FSB Advising Checklist - Orientation 2024
 
TỔNG ÔN TẬP THI VÀO LỚP 10 MÔN TIẾNG ANH NĂM HỌC 2023 - 2024 CÓ ĐÁP ÁN (NGỮ Â...
TỔNG ÔN TẬP THI VÀO LỚP 10 MÔN TIẾNG ANH NĂM HỌC 2023 - 2024 CÓ ĐÁP ÁN (NGỮ Â...TỔNG ÔN TẬP THI VÀO LỚP 10 MÔN TIẾNG ANH NĂM HỌC 2023 - 2024 CÓ ĐÁP ÁN (NGỮ Â...
TỔNG ÔN TẬP THI VÀO LỚP 10 MÔN TIẾNG ANH NĂM HỌC 2023 - 2024 CÓ ĐÁP ÁN (NGỮ Â...
 
Mixin Classes in Odoo 17 How to Extend Models Using Mixin Classes
Mixin Classes in Odoo 17  How to Extend Models Using Mixin ClassesMixin Classes in Odoo 17  How to Extend Models Using Mixin Classes
Mixin Classes in Odoo 17 How to Extend Models Using Mixin Classes
 
Mehran University Newsletter Vol-X, Issue-I, 2024
Mehran University Newsletter Vol-X, Issue-I, 2024Mehran University Newsletter Vol-X, Issue-I, 2024
Mehran University Newsletter Vol-X, Issue-I, 2024
 
Food safety_Challenges food safety laboratories_.pdf
Food safety_Challenges food safety laboratories_.pdfFood safety_Challenges food safety laboratories_.pdf
Food safety_Challenges food safety laboratories_.pdf
 

Situation Recognition from Multimodal Data Tutorial (ICME2016)

  • 1. /125 SITUATION RECOGNITION FROM MULTIMODAL DATA Vivek K. Singh1, Siripen Pongpaichet2, and Ramesh Jain2 1Rutgers University, 2University of California, Irvine
  • 3. /125 Course Outline 1) Concept recognition from Multimedia data (20 mins, Ramesh Jain) • Trends • Why situation recognition is different from object, event, scene recognition etc. 2) Situation recognition across multiple research domains (20 mins, Vivek Singh) • Situation Algebra, Situation Calculus, Robotics, … 3) Situation recognition (45 mins, Vivek Singh) • Situation modeling • Situation operators 4) Designing situation based applications (30 mins, Siripen Pongpaichet) • Motivation and essential requirements. • Application scenarios: Thailand flood, hurricane Sandy, city sensing, asthma relief 5) Future trends and open problems (20 mins, Ramesh Jain) • Future trends • Open problems for Multimedia research 6) Question and Answers (15 mins) 3
  • 4. /125 CONCEPT RECOGNITION FROM MULTIMEDIA DATA (20 mins, Ramesh Jain) 4
  • 5. /125 Introduction • Object, event, scene recognition • Trends • Situation recognition ( is different from object, event, scene recognition etc.) 5
  • 6. /125 Data, Information, Knowledge, Wisdom 6 Data is Essential. But, we are really interested in products: Information, Knowledge, and Wisdom.
  • 7. /125 What is Important in ‘Big Data’? Multimedia Realtime Uncertainty 7
  • 8. /125 The Grand Challenge Sense making from multimodal massive geo-social data- streams. 8
  • 9. /125 Fundamental Problem Connecting People to Resources effectively, efficiently, and promptly in given situations.
  • 10. /125 What is Cyber Space? Who invented it? Animals Machines Societies Published first in 1942 10
  • 11. /125 •Desired state (Goal) •System model and Control Signal (Actions) •Current State (using multimedia data) 11
  • 12. /125 Input Computed using System Model Feedback Output compared with desired goal Actual System Output Observed Continuously 12
  • 14. /125 Connecting People And Resources Social Life Networks Aggregation and Composition Situation Detection Alerts Queries Information 7/21/2016 14
  • 15. /125 Traditional Social Systems •Models of Systems were difficult to form. •Current State of the system could not be determined. •Real time ‘actions’ could not be implemented. 15
  • 16. /125 •Social models can be determined using warehouses of Big Data. •Social observations are now possible with little latency. •Actions could be targeted to precise sources. 16
  • 17. /125 EventShop : Global Situation Detection Predictive Situation Recognitio n Evolving Global Situation Predictive Personal Situation Recognitio n Personal EventShop Evolving Personal Situation Need- Resource Matcher Recommendation Engine PersonaDatabas e Resources Needs Data Ingestio n Wearable Sensors Calendar Location…. DataSources …. Data Ingestion and aggregatio n Database Systems Satellite Environmental Sensor Devices Social Network Internet of Things Actionable Information 17
  • 18. /125 Concept Recognition: Last Century 18 Environ ments Real world Objects Situations Activities SingleMedia SPACE TIME ScenesLocation aware Visual Objects Trajectories Visual Events Location unaware Static Dynamic Location aware Location unaware Static Dynamic Data = Text or Images or Video
  • 19. /125 Visual Concept Recognition: First research papers • 1963: Object Recognition [Lawrence + Roberts] • 1967: Scene Analysis [Guzman] • 1984: Trajectory detection [Ed Chang+ Kurz] • 1986: Event Recognition [Haynes + Jain] • 1988: Situation Recognition [Dickmanns] 1960 1970 1980 1990 2000 2010 Object Scene Trajectory Event Situation
  • 20. /125 Concept Recognition: This Century 20 Environ ments Real world Objects Situations Activities SPACE TIME Location aware Location unaware Static Dynamic HeterogeneousMedia Location aware Location unaware Static Dynamic Data is just Data. Medium and sources do not matter.
  • 21. /125 Concept recognition from multimedia data 21 SPACE TIME ScenesLocation aware Visual Objects Situations Visual Events Location unaware Static Dynamic HeterogeneousMedia SingleMedia 360 K 11.4K 3.4 K Location aware Location unaware Static Dynamic
  • 22. /125 Situation Recognition: Next Frontier • Data Abundance. • Big Opportunity for Multimedia • Time • Space • Multiple diverse data streams • Need new frameworks. 22
  • 23. /125 SITUATION RECOGNITION ACROSS MULTIPLE RESEARCH DOMAINS (20 mins, Vivek Singh) 23
  • 24. /125 Related Work: Data to Situations Area Combine hetero data Human sensors Data analytics Define situations Location aware Real-time streams Toolkits Situation Awareness X X o o X Situation Calculus X Web data mining o X X o X Social media mining o X X o X Multimedia Event detection X o o o Complex event processing/ Active DB X X o X GIS X o X X o Mashup toolkits (Y! pipes, ifttt) X X o X X X X X X X X X X X X X X This work X X X X X X X XX 24 o = partial support
  • 25. /125 Defining Situations: Situation Calculus • A situation s is the complete state of the universe at an instant of time. –McCarthy, 1968 • Snapshot of the world at a given time.- Reiter, 1991 • The set of necessary and sufficient world state descriptors for undertaking control decision. – Singh & Jain, 2009 25
  • 26. /125 Situation Calculus: Quick overview • enter(P1), startWork(P1) • enter(P1), exit(P1), enter(P1), startWork(P1), stopWork(P1), startWork(P1) - isInRoom(P1, s(k)) - isWorking(P1, s(k)) isInRoom(P1, s) ˄ ~isWorking(P1, s) → IncreaseMusicVolume() isInRoom(P1, s) 0 isWorking(P1, s) 01 1 Situation = Not events , nor sequence of events, but their assimilated descriptor
  • 27. /125 Problems with this approach • Scalability: • Listing all the rules • Frame problem - Specifying the non-effects • Assumes 100% confidence in events detected • Space is not a first class citizen. • Does not deal well with heterogeneous data 27
  • 28. /125 Situations • Multiple definitions • Situation awareness • Situation modeling • Situation detection • Situation calculus • Context based computing “the perception of elements in the environment within a volume of time and space, the comprehension of their meaning, and the projection of their status in the near future (Endsley, 1988)”. “knowing what is going on so you can figure out what to do” (Adam, 1993)”. “the complete state of the universe at an instant of time” (McCarthy, 1969) “a set of past contexts and/or actions of individual devices relevant to future device actions” ” (Wang,2004)”. “…extensive information about the environment to be collected from all sensors independent of their interface technology. Data is transformed into abstract symbols. A combination of symbols leads to representation of current situations…which can be detected”(Dietrich, 2003) “A situation is a set of contexts in the application over a period of time that affects future system behavior” (Yau, 2006)
  • 29. /125 Situations: commonalities 29 • Goal Based • Space-Time • Future Actions • Abstraction • Computationally Grounded Work Goal Based Space-Time Future Actions Abstraction Computationally Grounded McCarthy, 1968 X Barwise, 1971 X X Endsley, 1988 X X X X Sarter, 1991 o X Adam, 1993 X X Dominguez,1994 X X X X Smith, 1995 X o X X Steinberg, 1999 X X X o Jeannot, 2003 X Moray, 2004 o X Dietrich, 2004 X X Yau, 2006 X X X Dostal, 2007 o X Singh, 2009 X X X Merriam-Webster (accessed 2012) o This work (aim) X X X X X o = Partial support
  • 30. /125 Situation: Definition • Situation: An actionable abstraction of observed spatio-temporal characteristics. • e.g. flu epidemic, severe asthma threat, road congestion, wildfire, flash-mob 30 Goal Based Space-Time Future Actions Abstraction Computationally Grounded
  • 31. /125 Overall Framework: Motivating example 31 STT data Tweet: ‘Urrgh… sinus’ Loc: NYC, Date: 3rd Jun, 2011 Theme: Allergy Situation Detection User-Feedback ‘Please visit nearest CDC center at 4th St immediately’ Date: 3rd Jun, 2011 Aggregation, 1) Classification 2) Control action Operations Alert level = High
  • 32. /125 Applications • Healthcare • Alert me if there is a flu epidemic in my area • Telepresence: • Which camera feed to send out? • Business analysis: • Where is the most suitable place to open a new ‘iphone’ store ? • Weather • Alert me when the fall colors blossom in New England? • Daily living: • Which place (and at what time) is conducive for exercising? • Weather, climate, politics, traffic, … 32
  • 34. /125 A) Situation Modeling B) Situation Recognition C) Visualization, Personalization, and Alerts … STT Stream Emage Situation C1  v2 v3  v5 v6 @ ∏ Δ @ i) Visualization ii) Personalization + + Available resources iii) Alerts Personal context Personal ized situation Overall framework 34
  • 35. /125 A) Situation Modeling • Help domain experts externalize their internal models of situations of interest e.g. epidemic. • Building blocks: • Operators • Operands • Wizard: • A prescriptive approach for modeling situations using the operators and operands 35 Singh, Gao, Jain: Situation recognition: An evolving problem for heterogeneous dynamic big multimedia data, ACM Multimedia ‘12.
  • 36. /125 Growth rate (Flu reports) Feature Thresholds (0, 50) Data source Meta-data -Emage (#Reports) Representation level Twitter-Flu Building Blocks: Operands 36 • Knowledge or data driven building blocks
  • 37. /125 Building Blocks: Operators Δ Transform … Spatio-temporal window 37  Aggregate +  Classification Classification method @ Characterization Growth Rate = 125% Property required Pattern Matching 72% + Pattern ∏ Select + Mask Φ Learn Learning method {Features} {Situation} f f 1) Data into right representation 2) Analyze data to derive features 3) Use features to evaluate situations Supporting parameter(s) Data OutputOperator Type
  • 38. /125 Situation Modeling Get_components (Situation v){ 1) Identify output state space 2) Identify S-T bounds 3) Define component features: v=f(v1, …, vk) • If (type = imprecise) • identify learning data source, method 4) ForEach (feature vi) { If (atomic) • Identify Data source. • Type, URL, ST bounds • Identify highest Rep. level reqd. • Identify operations Else Get_components(vi) } } 38 v f1 v4  v2 v3 @ D1 Emage Δ D2 ∏ Emage Δ D3 Δ @ Emage D2 ∏ Emage Δ f2  v5 v6 <USA, 5 mins, 0.01x 0.01> ϵ { Low, Mid, High}
  • 39. /125 Epidemic Outbreaks Unusual Activity? Growth Rate  Current activity level Historical activity level  Emage (#reports ILI) Δ Twitter-Flu  Twitter.com <USA, 5 mins, 0.01x 0.01> Emage (Historical avg) Δ Twitter-Avg DB, <USA, 5 mins, 0.01x 0.01> Δ Twitter-Flu Emage (#reports ILI) Twitter.com <USA, 5 mins, 0.01x 0.01> ϵ {Low, mid, high}, <USA, 5 mins, 0.01x 0.01> Growing Unusual activity  1)Model Emage (#reports ILI) Δ Twitter-Flu Emage (population) Δ CSV- Population  π Twitter.com <USA, 5 mins, 0.01x 0.01> Census.gov, <USA, 5 mins, 0.01x 0.01> 2) Revise Subtract Subtract Multiply Classification: Thresh (30,70) Normalize [0,100] 3) Instantiate 39
  • 40. /125 Level 1: Unified representation (STT Data) Level 3: Symbolic rep. (Situations) Properties Properties Properties Level 0: Raw data streams e.g. tweets, cameras, traffic, weather, … Level 2: Aggregation (Emage) … STT Stream Emage Situation B) Situation evaluation: Workflow 40 Operations
  • 41. /125 Data Representation • E-mage • Visualization • Spatio temporal data representation • Data analysis using media processing operators (e.g. segmentation, background subtraction, convolution) 41
  • 42. /125 Data Model • Spatio-temporal element • stel = [s-t-coord, theme(s), value(s), pointer(s)] • E-mage • g = (x, {(tm, v(x))}|xϵ X = R2 , tm ϵ θ, and v(x) ϵ V = N) • Temporal E-mage Set • TES= {(t1, g1), ..., (tn, gn)}, • Temporal Pixel Set • TPS = {(t1, p1), ..., (tn, pn)},
  • 43. /125 Situation Recognition Algebra Singh, Gao, Jain: Social Pixels: Genesis and Evaluation, ACM Multimedia ‘10. 43 S. No Operator Input Output 1 Filter ∏ Temporal E-mage Stream Temporal E-mage Stream 2 Aggregation  K*Temporal E-mage Stream Temporal E-mage Stream 3 Classification  Temporal E-mage Stream Temporal E-mage Stream 4 Characterization : @  Spatial  Temporal  Temporal E-mage Stream  Temporal Pixel Stream  Temporal Pixel Stream  Temporal Pixel Stream 5 Pattern Matching   Spatial  Temporal  Temporal E-mage Stream  Temporal Pixel Stream  Temporal Pixel Stream  Temporal Pixel Stream
  • 45. /125 Implementation and results • Twitter feeds • Geo-coding user home location • Loops of location based queries for different terms • Over 100 million tweets using ‘Spritzer’ stream (since Jun 2009), and the higher rate ‘Gardenhose’ stream since Nov, 2009. • Flickr feeds • API • Tags, RGB values from >800K images
  • 46. /125 Testing Data Representation + Algebra • Applications • Business analytics • Political event analytics • Seasonal characteristics • Data • Twitter feeds archive • Loops of location based queries for different terms • Over 100 million tweets using ‘Spritzer’/ ‘Gardenhose’ APIs • Flickr feeds • API: Tags, RGB values from >800K images • Implementation • Matlab + Java + Python 46
  • 47. /125 Sample Queries • Select E-mages of USA for theme ‘Obama’. • ∏spatial(region=[24,-125],[24,-65]) (TEStheme=Obama) • Identify three clusters for each E-mage above. • kmeans(3) (∏spatial(region=[24,-125],[24,-65])(TEStheme=Obama)) • Show me the cluster with most interest in ‘Obama’. • ∏value(v=1) (kmeans(n=3) (∏spatial(region=[24,-125],[24,-65]) (TEStheme=Obama))) • Show me the speed for high interest cluster in ‘Katrina’ emages • @speed(@epicenter(∏value(v=1) (kmeans(n=3) (∏spatial(region=[24,-125],[24,-65]) (TEStheme=Katrina))))) • How similar is pattern above to ‘exponential increase’? • exp-increase(@speed(@epicenter (∏value(v=1) (kmeans(n=3) (∏spatial(region=[24,-125],[24,-65]) (TEStheme=Katrina)))) 47
  • 48. /12548 AT&T retail locations AT&T total catchment area iPhone theme based e-mage, Jun 2 Aggregate interest Under-served interest areas - Subtract Decision Best Location is at Geocode [39, - 122] , just north of Bay Area, CA @Spatial.Max <geoname> <name>College City</name> <lat>39.0057303</lat> <lng>-122.0094129</lng> <geonameId>5338600</geonameId> <countryCode>US</countryCode> <countryName>United States</countryName> <fcl>P</fcl> <fcode>PPL</fcode> <fclName>city, village,...</fclName> <fcodeName>populated place</fcodeName> <population/> <distance>1.0332</distance> </geoname> + Add to Jun 15, 2009 Convolution . * Store catchment area Convolution. * Store catchment area
  • 49. /125 Flickr Social Emages • Jan – Dec 2009
  • 50. /125 Seasonal characteristics analysis • Fall colors in New England • Show me the difference between red and green colors for New England region, as it varies throughout the year. • subtract(@spatial(sum)(πspatial(R=[(40,-76), (44,-71)]) (TEStheme=Red)), @spatial(sum)(πspatial(R=[(40,-76), (44,-71)])(TEStheme=Green))) 50 Jan 0 Dec
  • 51. /125 Year average Peak of green At [35, -84], at the junction of Chattahoochee National Forest, Nantahala National Forest, Cherokee National Forest and Great Smoky Mountains National Park
  • 53. /125 FraPPE: a vocabulary to represent heterogeneous spatio-temporal data to support visual analytics Marco Balduini, Emanuele Della Valle ISWC 2015 – Data Sets and Ontologies Slides adapted from: http://www.slideshare.net/MarcoBalduini/frappe-a-vocabulary-to-represent-heterogeneous-spatiotemporal-data-to-support-visual-analytics
  • 54. /125 Proposed Approach 54 21/07/2016 Re-use consolidated concepts from • geo-spatial vocabularies • time related vocabularies • provenance vocabularies Model visual analytics concepts * • pixels • frame Fill the gap between heterogeneous geo-spatial time-varying data and visual analytics concepts to create actionable information and ease the decision making processes of final users * Singh, V.K., Gao, M., Jain, R.: Social pixels: genesis and evaluation (ACM MM 2010)
  • 61. /125 City Sensing listens to the pulse of Milano Design Week on April 9th, 2014 61
  • 62. /125 Tweeting Cameras 62 Slides courtesy: Yuhui Wang, Francesco Gelli, and Mohan Kankanhalli Adapted from “Tweeting Cameras for Event Detection” in Proc. WWW Conference 2013.
  • 63. /125 Physical & Social Sensors Fusion For Situation Awareness Physical Sensors Social Sensors
  • 64. /125 Real-world Events Hispanic ParadeCBGB Musical Festival Columbus Day Parade twtw tw tw tw tw twtw tw tw tw tw twtw tw tw tw tw twtw tw tw tw tw Historic TweetsRecent Tweets Event time and location Retrieve recent tweets Retrieve historic tweets
  • 65. /125 Probabilistic Spatio-Temporal Data • Definition (PST: Probabilistic Spatio-Temporal Data) The fundamental building block for low-level concept representation is the probabilistic spatio-temporal element “pst”. • pst = [loc, temp, label, prob, pointer] (1) where • loc = [lat, lon] represents the geo-location – latitude and longitude – of the camera location. • temp stores the time information of captured data. • label represents semantic concept such as car, human,crowd, parade, etc., detected in the stream. Generally, these concepts express low-level abstraction of information which could be semi-reliably detected by existing detectors or classifiers. • prob is the confidence value in [0,1] representing the output of a concept detector as a probability value. • pointer points to actual raw data stream. 65
  • 70. /125 Fusing Sensor Cmage with Social Cmage Sensor Cmage (concept: “people marching”) Social Cmage (concept: “MillionsMarchNYC”) Fused Cmage
  • 71. /125 DESIGNING SITUATION BASED APPLICATIONS (30 mins, Siripen Pongpaichet) 71
  • 72. /125 Outline • EventShop System Requirements • EventShop System Architecture • Demo • Building Applications using EventShop • Conclusion 72
  • 73. /125 EventShop Requirement 73 Granulari- ties Heterogen eous Model Prediction Users Open- Source Storage Generic Streams Support fast data flow Handle heterogeneous types of data streams Efficiently aggregate data at different granularities Provide storage system to archive both data input and system output. Create situation model and provide actionable information Generic computational platform for situation recognition Open-source software User friendly and interactive interface Contain predictive component
  • 74. /125 EventShop Architecture 74 Alert/OutputData Ingestor Data Source Parser Data Adapter Emage Generator (+resolution mapper) Query Processing EventShop Storage Query Parser Query Rewriter Emage Stream Processing Action Parser Situation Emage Visualization (Dashboard) Actuator Communication Event Property & Other Information (e.g., spatio-temporal pattern) ᴨ ᴨ µ Data Access Manager Live Stream Archived Stream Situation Stream Physical Data Source (e.g., sensor streams, geo-image streams) Logical Data Source (e.g., preprocessing data streams, social media streams) Raw Event REST API Services Data Source / Query / Alerts / STT-Emage EventShop UI External AppsVisual Analytics InterfaceAPIProcessingLayerStorageLayer
  • 75. /125 EventShop UI 75 Save Query Reset Query Create Query Pollen Tweets_ Asthma Available Available 350 AQI Available357 361 Grouping Stopped35 Asthma_ Risk Stopped36 Asthma_ Interpolate Stopped37 Asthma_ Interpolate Stopped37 Asthma_ Stopped37 Query Graph Filter Group Aggregation Spatial Pattern Temporal Pattern Spatial Char Temporal Char Redraw ds361 ds350 ds357 F1 Q1 F2 F3 Q2 Q3 A1 Q4 G1 Q5 http://eventshop.ics.uci.edu:8080/eventshoplinux/
  • 78. /125 Building applications using EventShop S.No Application Data Used Application deployed? Scale Data modalities Operators used 1 Wildfire detection in California Real Yes Macro Satellite data, Google insights F, A, Ch 2 Hurricane monitoring Simulated No Macro n/a F, A, Ch, P 3 Flu epidemic surveillance Real No Macro Twitter, Census F, A, C 4 Allergy recommendation Real Yes Macro Twitter, Air Quality, Pollen Count F, A, C 5 Asthma management Real Yes Macro, Personalized alerts In situ sensors, Satellite data, Asthma Tracking F, I, Pr 6. Thailand flood mitigation Real Yes Macro, Personalized alerts KML F, A, C 7. Photos as Micro- Reports Real Yes Macro Flickr F, Cl 8. Trash management Real & Simulated In progress Macro Trash sensors, micro-reports F, A, Pr, 78
  • 80. /125 Asthma Management Application 80 (1) Macro Situation Macro Data Streams (3) Situation- Action Rules Sensor streams Social media Geo-temporal data Personal Data Streams (2) Personal Situation Behavioral streams Profile + Preferences
  • 81. /125 Asthma Risk Estimation 81 Traffic Flow Aerosol Concentration PM2.5 CMAQ Model PM2.5 Concentration *visualize data on Feb 12th, 2008 Mengfan Tang, Pranav Agrawal, Siripen Pongpaichet, Ramesh Jain: Geospatial interpolation analytics for data streams in EventShop. ICME 2015 Spectral Spatial Gaussian Process (SSGP)
  • 82. /125 Experiment Results 82 Data Model PMSE MAPE Single Data Source CMAQ - 1.0619 27.2873 CMAQ LR 0.9586 27.1077 Stations Kriging 0.9077 22.9672 CMAQ SSGP 0.3468 14.2727 Multiple Data Sources ALL SGP 0.3006 13.5109 ALL SSGP 0.2858 13.1087 CMAQ Kriging SSGP
  • 83. /125 Asthma Risk Estimator Model and Result 83 Asthma Hospitalization Ground Truth FILTER LOC=CA FILTER LOC=CA AGG FUNC=AVG GROUP THRESHOLD Asthma Risk Area without Interpolation GROUP THRESHOLD Asthma Risk Area with interpolation AGG FUNC=AVG PM2.5 Concentration From Stations Interpolated PM2.5 using SSGP Pollen Ozone AQI
  • 84. /125 A GRAPH BASED MULTIMODAL GEOPATIAL INTERPOLATION FRAMEWORK Mengfan Tang1, Pranav Agrawal1, Feiping Nie2, Siripen Pongpaichet1, Ramesh Jain1 1University of California, Irvine, USA 2Northwestern Polytechnical University, China Tuesday July 12th, 2016 at 5PM Room: Cascade I, Special Session: Multimedia Cloud Computing and Big Data
  • 87. /125 Disruption: PHOTOS as Information Smartphone camera captures EVENTS
  • 88. /125 • Truthfulness, • Accuracy, • Objectivity, • Fairness and Public accountability Reports of Events from Journalists Seek Truth and Report it as Fully as Possible
  • 89. /125 Reports of Events from Citizens
  • 90. /125 FASTSubjective EASY Noisy LATES T Ambiguous were SO yesterday!Micro-Blogs Multimedia Micro-Reports
  • 91. /125 Compelling Universal Objective Spontaneous Multimedia Micro-Reports (MMRs) are now and future
  • 92. /125 Capturing and Reporting events using multimedia such as photos, videos, sensors, and texts Converting multimedia data to multimedia micro-reports using MediaJSON Integrating multimedia micro-reports with other data sources for situation recognition, trend analysis, and culture analytics. Emerging opportunities for numerous apps: smart city, public health, emergency rescue What are the challenges?
  • 93. /125 Capturing and Reporting Events with Krumbs SDK https://krumbs.net/ What: Objects Who: People When: Events Where: Location Why: Intent/Emotions How: Photo and audio
  • 94. /125 Capturing and Reporting Events with Krumbs SDK https://krumbs.net/
  • 96. /125 Converting Multimedia Data into MMR {"micro_reports":[{ "where":{ "geo_location":{ "latitude":32.90233332316081, "longitude":-117.2441166718801}, "when":{ "start_time":"Jun 14, 2009 11:25:19 AM", "end_time":"Jun 14, 2009 11:25:19 AM", "time_zone":"America/Los_Angeles"}, "what":[{ "concept_name":"people", "confidence":0.9836078882217407, "visual_concept_provider":"CLARIFAI"}, … { "concept_name":"food", "confidence":0.8526291847229004, "visual_concept_provider":"CLARIFAI"}] , "tag":”#niceday #summer", "source":{"default_src":"https://….jpg"}}, "sub_event":[], "why":[]}, …]} Photo What Where When Who Why Sound MediaJSON
  • 98. /125 Number of photos in London per day
  • 100. /125 • Year of interest: year 2008 and 2012 • Training locations: Beijing (China) • Testing location: London (UK) • Create “Olympic Games” Event Model from “BAG of Visual Concepts” Detecting Olympic Games Olympic Games = {basketball, court game, gymnastics, people, sport, stadium, swim, tennis}
  • 101. /125 Evolving Photo Concepts in Beijing Event model of “Olympic Game”
  • 102. /125 Detecting London Olympic Games Summer Olympic Game in July and August Paralympic Games in September
  • 103. /125 • Temporal range: 1 year from July 2011 to June 2012 • Location: Thailand Detecting Emergency Situations City flood = {outdoor, water, road, car}
  • 104. /125 Photos from City Flood Cluster November 5, 2011
  • 105. /125 Smart City Project in DC 105 U.S. Presidential Inauguration in DC Earth Day Concert Cherry Blossom Festival
  • 106. /125 Integrating MMR with other data sources for Situation Recognition (In progress) h t t p : / / s m a r t c i t i e s i n n o v a t i o n . c o m /
  • 107. /125 Trash Fill Level Situation in DC 107
  • 108. /125 0 35 50 90 0 20 40 60 80 100 7:30 8:00 8:30 9:00 9:30 10:00 10:30 Real-Time Fill Level Situations at a given location of an event Prediction based on Events History Events Data Real-Time Trash Fill Level Situation 20 42 Now Predicted Trash Fill Level in 30 minutes at a given location 78 99 30 minutes 10 30 40 70 90 100 20 0 20 40 60 80 100 120 7:30 8:00 8:30 9:00 9:30 10:00 10:30 Projected Trash Fill Level at a given location based on Event History
  • 109. /125 Conclusion • EventShop Architecture • Situation-based Applications using EventShop • Online Service: http://sln.ics.uci.edu:8085/eventshoplinux • Open Source: http://dabuntu.github.io/es/ • For more information about EventShop including tutorial videos, presentations and publications, please visit my home page http://www.ics.uci.edu/~spongpai Siripen Pongpaichet (spongpai@uci.edu) 109
  • 110. /125 FUTURE TRENDS AND OPEN PROBLEMS (20 mins, Ramesh Jain) 110
  • 111. /125 Future Trends • Future trends • Open problems for Multimedia research 111
  • 112. /125 This century is different from the last. Should we think differently???
  • 113. /125In 20th century, we tolerated photos in our textual documents. In 21st century, you create visual documents that tolerate text.
  • 114. /125 Major Disruption in Photos: From Memories to Information Sources. Photos are the most compelling source of information.
  • 116. /125 We are immersed in Big Data. Multimedia Realtime Uncertainty 116
  • 118. /125 Data as a Platform. • Multi-modal • Multimedia has to become multimodal • Data Streams • Important things – Situation recognition • Real time action for 118 Connecting People to Resources effectively, efficiently, and promptly in given situations.
  • 119. /125 Contact Information • Vivek Singh, Rutgers University • Vivek.k.singh@Rutgers.edu • • Siripen Pongpaichet • spongpai@ics.uci.edu • Ramesh Jain • jain@ics.uci.edu 119
  • 120. /125 Useful links • Copies of publications • http://wp.comminfo.rutgers.edu/vsingh/publication/ • Today’s Slides: • https://dl.dropboxusercontent.com/u/5887580/Tutorial_SituationRecog. pdf • EventShop • Online Service: http://sln.ics.uci.edu:8085/eventshoplinux • Open Source: http://dabuntu.github.io/es/ • For more information about EventShop including tutorial videos, presentations and publications: http://www.ics.uci.edu/~spongpai • Related Projects • Tweeting Cameras: https://sites.google.com/site/fredyuhuiwang/ • Frappe: http://www.streamreasoning.org/live/festivalcomunicazione2014/ 120

Editor's Notes

  1. Bring Predictive here.
  2. A real part of the world is represented by the map in the figure. The data geo-spatial time-varying is too complex to be used for a visualization.
  3. A GRID, composed by 4 CELLS, is placed between the map and the film that will capture the actual reflection of the reality. The GRID is needed to aggregate the date over space
  4. A film capture a FRAME, made by 4 PIXEL, that represents the temporal representation of the reality Using the PIXEL and the FRAME FraPPE aggregates the data over time.
  5. Once something happen in the PLACE A (geo-laceted place in a CELL and in a GRID) the reality, for example the pick-up event of a taxi ride, the related EVENT is impressed over the film in the corresponding PIXEL and FRAME.
  6. The reality is mediated again with the 4X4 GRID
  7. A new FRAME is impressed on the film, representing a new reflection of the reality with a new EVENT that is happening in PLACE B and I represented on the FRAME by the EVENT b
  8. We are trying to fuse physical sensors (e.g. CCTV cameras with social sensors (Twitter tweets ) together to have a better understanding of ongoing situations
  9. Red dots are fixed camera at different locations in Manhattan, tweeting about different concept (expressed by color intensity, redder means high confidence of the concept) world clouds are generated by mining tweets content using tf-idf (we put geo tagged human tweets round a camera location during event time of event day into one document, and tweets of same location, same time span but past days into other documents; words with high tfidf mean they are discussed more frequently during event hours of event days then same hours but past days.)
  10. Red dots are fixed camera at different locations in Manhattan, tweeting about crowdedness (expressed by color intensity, redder means more crowded)
  11. Green dots are human tweets containing hashtags (#MillionMarchNYC, #BlackLivesMatter) at different locations in Manhattan. Left top is the their marching route ground truth.
  12. Red dots are fixed camera at different locations in Manhattan, tweeting about crowdedness (expressed by color intensity, redder means more crowded) Green dots are human tweets containing hashtags (#MillionMarchNYC, #BlackLivesMatter) at different locations in Manhattan. Left top is the their marching route ground truth. Combining camera tweets with human tweets, world clouds are generated by mining tweets content using tf-idf (we put geo tagged human tweets round a camera location during event time of event day into one document, and tweets of same location, same time span but past days into other documents; words with high tfidf mean they are discussed more frequently during event hours of event days then same hours but past days.) As can be seen, though cameras at different places tweet about “crowdedness”, they mean different events (mid-Manhattan is “Santacon Event” while lower Manhattan is “MillionMarchNYC” protest event )
  13. From camera feeds to concept image (Cmage) : Gaussian Process could be applied on such data representation to infer situations in place without sensors
  14. Fusing sensor cmage with social cmage for “MillionsMarchNYC” protest events Sensor cmage intensity represents confidence values of “people marching” concept extracted from CCTV camera images Social cmage intensity is tf-idf values of term “MillionMarchNYC” calculated from Twitter tweets In fused image, noise has been removed and saliency of events has been enhanced
  15. Support fast data flow Handle heterogeneous types of data streams Efficiently aggregate data at different granularities Provide storage system to archive both data input and system output. Create situation model and provide actionable information Contain predictive component Generic computational platform for situation recognition Open-source software User friendly and interactive interface
  16. Traffic: loop detectors & traffic monitoring sparse in space, dense in time – from caltrans performance measurement system Aerosol Concentratation: satellite data, sparse in space and time – from MODIS Aerosol optical depth PM2.5 CMAQ: Chemical transportation model, sparse in space and time – from community modeling and analysis system PM2.5 Concentration station: air measurement stations, limited # of stations – US EPA (environmental projecting agency) air quality system
  17. multimedia research on object recognition, scene recognition, event recognition - on photos summarization, Browse through photos on different dimension. Space-time-people-event..
  18. We Capture and Share Moments -- Really Events What are these natures of photos lease to new opportunities. Trillion photos have been uploaded last year. 7 billions people in the world. 150 photos have been uploaded by individual person.
  19. Now. Let’s talk about REPORTS! For century, in journalism philosophy, the reports MUST have these characteristics. Truthfulness, accuracy, objectivity, and fairness. All in all, the goal is to “seek the truth and report it as fully as possible” without reporters’ own opinion.
  20. How people report the events? Through all these social media; twitter, facebook, snapchat, instagram. For example, during ucla shooting last week, people tweeted about it, took some pictures and videos as they were witnessing that situations.
  21. Pro Became very powerful very fast Easy to use – like SMS. Aggregated for trend analysis, analytics, visualization of evolving situations. Con Missing/misleading context – ambiguous – poor context – lots of research on that sentiment analysis Noise – limited number of character, all topics are mixed together, Hashtag # for addressing a theme, however, this doesn’t solve the problem. Subjective These disadvantages violate all journalism philosophy !!! Great concept but has limitations. New technology can overcome these limitations. We need a new mechanism to handle this new type of report – micro-report What is the successful story of focused micro-reports- WAZE This multimedia micro-reports concept is inspired by Focused MicroBlogs first introduced by Ramesh Jain. http://ngs.ics.uci.edu/focused-microblogs-fmbs-going-beyond-twitter/
  22. Compelling (context around the photos) Universal (independent of Language) Objective (not opinion) Spontaneous (share immediately) ---- Compelling: Micro-reports should effortlessly capture the moment with rich context along with compelling photo, audio, video and/or other signals. As are well known photos, videos, and audio which make a report of an experience more compelling and more informative. Universal: Micro-reports need to have a universal language, that is beyond literacy and geographical boundaries. Photos are universal language. Objective: Micro-reports should emphasize on facts and not subjective opinions of people and their personal atti- tude toward the facts. Opinions are important, but should appear explicitly as opinions. Objective and subjective components in a report should be clear. Spontaneous: Micro-reports should be shareable immediately. If creating a report requires thinking and typing a message, fewer people will use it during the event. A report submitted later relies on memory and may not have correct contextual information.
  23. Event-driven operation!!! Reports sent by IoTs
  24. For each day, the probabilities for the same concept is aggregated. Figure 6 gives an example of how four concepts, people, running, sport and swim evolve during the time of interest. Let’s first look at the dynamic change in the ‘people’ concept, we observe that London is less crowded in winter season and attracts more people in summer and autumn. London usually hosts many kinds of sport events, such as swimming, and running, which also can been detected in this figure. However, single data stream (‘people’ theme) finds itself hard to detect situations. Social events and trends are usually detected by the concurrences of photo’s visual concepts.
  25. bounding box of London (-0.489,51.28 0.236,51.686)
  26. Figure 5 shows the evolving concepts in Beijing in the year of 2008. As we know, Beijing Olympic Games was held in Beijing in the summer of 2008. This event is clearly reflected in Figure 5 where the number of these ten concepts ‘basketball’,‘court game’,‘gymnastics’,‘people’,‘sport’,‘stadium’,‘swim’ and ‘tennis’ reaches peak at the same time of Olympic Games.
  27. Summer Olympics in July and August and Paralympic Games in September in
  28. From the historical data and human knowledge, these concepts, ‘outdoor’, ‘wa- ter’, ‘road’, and ‘car’, are highly occurred together during a city flood
  29. Photos from one of the cluster in Thailand on November 5, 2011. The length of edges in the graph represents a similarity between a flood event model (rectangle) and a photo (circle) considering its visual concepts. The shorter length means the higher similarity. Most of the photos related to flood event can be captured by our framework.
  30. Prob/Opportunity Slide
  31. Prob/Opportunity Slide