Computer vision brings great potential for automation, while also being associated with fears of violations of privacy. Let us discover how we can reap the benefits while mitigating the risk.
AI+A11Y 11MAY2024 HYDERBAD GAAD 2024 - HelloA11Y (11 May 2024)
Computer Vision - The New Renaissance or 1983?
1. Computer Vision:
The New Renaissance or 1983*?
Bogdan BOCȘE
Managing Partner VisageCloud
#AwakeFestival #Teleki #FeedYourMind
2. Who am I?
Bogdan BOCȘE
Managing Partner of VisageCloud
Founding Member of DeepVISS
Geek with a passion for
#AI #Math #MachineLearning
#ComputerVision #IoT #EdgeComputing
#AugmentedIntelligence
9. The Virtuous Circle of Computer Vision
Ubiquitous Data
Collecting, transmitting, processing
data is becoming cheaper
Increasing processing
power
Parallel processing is becoming cheaper,
more wildly-spread and energy efficient.
Computer Vision
Going
Mainstream
New Models and Algorithms
Getting better accuracy at higher speed with similar
cost.
11. Ultimately, Money Talks
“The Overall Computer Vision Market is Expected to Be Valued at USD 11.94 Billion in 2018
and is Likely to Reach USD 17.38 Billion By 2023, at a CAGR of 7.80% Between 2018 and
2023.”
Markets and Markets, 2018
“Facial Recognition Market is expected to garner $9.6 billion by 2022, registering a CAGR of
21.3% during the forecast period 2016-2022.”
Allied Market Research, 2016
13. Design Principles
1
Transparency
of Legitim
ate
Use
Technical m
echanism
s
for preventing
abuse.
2
Privacy
by
Design
Support data
retention
policy, support
m
arking
inform
ation
as
private.
3
Context-aw
areness. H
um
an-decision.
Correlate
events
from
several
stream
s
in
order to
produce
second
order events.
Rely
on
hum
an
for approving
actions.
14.
15. Deep Vision Interoperability Specification Standard
● Standard for integrating computer vision components
● Defines structured semantics of what computers can see
● Allows several components to work together
● Simplifies the development of computer vision applications
● Allow marking data as private
18. How can you contribute?
1. Spread the word @DeepVISS
2. Become a DeepVISS partner on https://deepviss.org
3. Review published drafts of standards
https://github.com/deepviss/deepviss-server-example
4. Give feedback on how DeepVISS can fit your use case
21. Multi-tier Architecture in Computer Vision
AI and Algorithms
1. Computer Vision
2. Machine Learning
3. CNNs
3
Storage, Retrieval, Search
1. Persistence
2. Retention Policy
3. Search
2
API, Integration and UI
1. Data Acquisition
2. APIs
3. User Interface
1
22. Parking management
& payment04
● Scan license plate and face at entrance
● Pay based on face
● Exit parking lot based on face or license plate
Abnormal event
detection03
● Detect non-specific anomalies
● Notify human operator for review
Loitering Detection02
● Notice an individual with a prolonged presence
in the same location
Detect Luggage
Abandoned by Owner01
● Establish ownership of luggage
● Issue alert when owner(s) exits perimeter
More Use Cases
23. Advantages of DeepVISS
Composability
Merge several solutions to
achieve desired practical and
commercial objective
● Face Recognition
● Emotion Classification
● Tracking Cars
● License Plate
Recognition
● Anomaly Detection
Structured Output, Effortlessly
Output defined in Open API
(similar to JSON Schema)
Instant build in any language
(C, Java, Python, Go)
Clear semantics around key
concepts:
● Detection
● Features
● Attributes
● Geometry
● Landmarks
Reusable Components
Save time and effort, without
having to deal with IT
components:
● Scalable Database
● API
● Data Acquisition
● Rate Limiting
● User Interface