o Moving objects and media data computing
- Spatial Big Data Computing
- Moving Objects and Media Data Computing
- Integrating Spatial Media and Deep Learning
Artificial intelligence in the post-deep learning era
Moving objects media data computing(2019)
1. Moving Objects and Media
Data Computing
(이동체 미디어 데이터 컴퓨팅 연구 동향)
2019. 07. 14
Kwang Woo NAM
kwnam@kunsan.ac.kr
Kunsan National University
이 연구는 국토해양부의 국토해양부의 국토정보연구 프로그램(14NSIP‐B080144‐01)과
한국연구재단 중견연구사업(No.2018R1A2B6007982), 한국국토정보공사의 산학협력 R&D 지원
사업에 의해 지원되었습니다. 2019.07. 15 Pitta Soft
http://infolab.kunsan.ac.kr
2. 2
Who
Professor/Kunsan National University
‐ KSNU, School of Computer, Information, and
Communications Engineering, Professor (2004‐now)
‐ Future Vehicle R&D Training Project (2017‐now)
미래형자동차 R&D 전문인력양성사업단(2017‐) : { KEA, 군산대, 한양대, 충북대, 인하대, 부품연 }
‐ U. of Minnesota, Visiting Scholar(2015‐2016)
‐ ETRI, Telematics Division, Senior Researcher(2001‐2004)
Projects
• 한국연구재단, 지오미디어 스트림 데이터 컴퓨팅 및 공간 딥러닝 질의 지원(18‐21) : Deep Learning+Spatial Media
• 한국국토정보공사, 도시상태 탐지를 위한 도시 환경 GeoAI 허브 기반 기술 개발(18‐19) : Deep Learning+Spatial
• 국토해양부, 공간정보 S/W활용을 위한 오픈소스 가공기술 개발(14‐19) : PostTrajectory, PostGeoMedia
• NRF, 소셜 미디어 스트림 기반 공간 지식의 연속 마이닝(13‐16): SKSpark, SpatialTajo
• ETRI, 시멘틱 공간정보 서비스 프레임워크 및 구조 분석(10)
• 국토해양부, U‐GIS 국토정보 컨텐츠 서비스 요소 기술 연구(07‐11)
• NRF, geoSensor 센서 스트림 기반 점진적 온라인 패턴 마이닝 기법 연구(09‐12)
• NRF, GeoSensor Network를 위한 효율적인 공간 센서 스트림 죠인 및 집계처리 기법 연구(06‐07)
• NRF, 상황정보 기반의 Networked GeoVision 기술 개발(08‐11) : GeoVIsion
3. Introduction
Spatial Data
Spatial Media and Contents
Spatial Information
Spatial Data Spectrum
2D Vector
RS Data
2.5D Data DEM Data
LBS/TelematicsGeoERPWeb GIS
MultiSensor
Smart Phone
Real 3D Data 360 Panorama
Video+GPS
(blackbox)
Game,Edu,…
Closed Box(Public Sector)
Open Box( Enterprise Sector)
Shared Universe(People Sector)
Sony Camera
(GPS+Pano) iWatch
Google Glass
GoPro
• Evolutions of Spatial Data/Information
4. Introduction
• 일반적인 공간좌표 및 POI 등의 공간 정보와 웹 문서, 도서, 사진,
동영상, 음성 등의 미디어가 결합된 공간정보 컨텐츠들의 등장
• 예) GoPro, PointCloud, 차량 블랙박스, 360 파노라마, 구글 글래스
GeoTagged Photo(Panoramio) 360 Panorama(StreetView)
Car Blackbox Video PointCloud(LiDAR) by Self Driving Cars
5. 5
• Converging into Spatial Data
Introduction
GNSS - 무선데이터통신망을
이용한 정밀위치정보제공
- 차량항법, 교통정보 제공
GIS SIIS
ITS
- 정밀 기준점 위치정보에
의한 영상정보 정확도 향상
- GPS 기지국 구축적지 선정
- 차량용 네비게이션 정보제공
- 교통주제도 구축 및 계량적
분석 기술제공
- 교통시설물 관리 및 IT기반
운영기술 제공
- 대용량 공간정보
자동수정 갱신
- 영상지도 제작
- 측량기술 및 관련기술
(위성삼각측량, WGS84)제공
- 실시간 Mobile GIS응용기술
- 영상기반 교통정보
수집기술제공
- 고 정밀 영상검지
시스템 구축
출처 : ETRI
19. 19
• PostGIS
• PostgreSQL Spatial Extension
• SQ1: Retrieve all roads crossing Highway #1
• SELECT gid, name FROM bc_roads
WHERE ST_Crosses(geom, ST_GeomFromText($Highway#1, 3005) );
• SQ2: Retrieve total length of all roads in the seoulRoad tables
• SELECT Sum( ST_Length( geom ) ) / 1000 AS km_roads
FROM seoulRoad;
Spatial Database Systems
ST_Intersects(G1,G2)
ST_Contains(G1,G2)
ST_Within(G1,G2)
ST_Touches(G1,G2)
ST_DWithin(G1,G2,D)
경부고속도로
20. 20
• pgRouting
• Routing function extension to PostGIS
• SQ3 : Calculate the shortest path between A and B
• Quebec Snow Mobile Trail Map
• http://fcmq.viaexplora.com/carte‐motoneige/indexen.html
Spatial Database Systems
SELECT * FROM pgr_dijkstra('
SELECT gid AS id,
source,
target,
cost_s / 3600 * 100 AS cost,
reverse_cost_s / 3600 * 100 AS
reverse_cost
FROM ways',
13009, 3986);
31. 31
• GeoHash as a Row Key
• Pros.
① Easy to calculate
② Prefix can be used to find nearest neighbors
• Cons.
• Accuracy and Boundary Problem
Spatial Big Data : Micro Blogs
33. 33
• Finding Home/Visitor
• Finding Spatial Words
Spatial Big Data : Micro Blogs
(mid, userid, x, y,
time, {word,…}) (<userid,geoid>, mid)
Map Shuffle
(<userid,geoid>, {mid,
mid, …})
Reduce
(<userid,geoid>, n)
(mid, userid, x, y,
time, {word,…}) (<geoid, word>, mid)
(<geoid, word>, mid)
(<geoid, word>, mid)
Map Shuffle
(<geoid, word>,{mid,
mid, …})
Reduce
(<geoid, word>, n)
34. 34
• SpatialHadoop
• Hadoop Extension with Spatial Operators
• Components
• MapReduce‐based Operations
• Two‐level Spatial Indexes
• Global index : Grid File
• Local index : R‐tree
Spatial Big Data : Systems
Ahmed Eldawy, Mohamed F. Mokbel:
SpatialHadoop: A MapReduce framework for spatial data. ICDE 2015: 1352‐1363
35. 35
• Hadoop‐GIS
• HiveSP: Spatial Data Warehouse System
• Spatial Partitioning
• Spatial Query Engine
• Two‐level Spatial Indexes
Spatial Big Data : Systems
Ablimit Aji, Fusheng Wang, Hoang Vo, Rubao Lee, Qiaoling Liu, Xiaodong Zhang, Joel H. Saltz:
Hadoop‐GIS: A High Performance Spatial Data Warehousing System over MapReduce. PVLDB 6(11): 1009‐1020 (2013)
36. 36
• Spark : In‐memory Computing
• GeoSpark
• Spatial Extension of Aparch Spark
• SpatialRDD with Spatial Data Types
• Memory‐level Spatial Local Index
Spatial Big Data : Systems
Jia Yu, Jinxuan Wu, Mohamed Sarwat:
GeoSpark: a cluster computing framework for processing large‐scale spatial data. SIGSPATIAL/GIS 2015: 70:1‐70:4
37. 37
• SpatialSpark
• Spatial Extension of Aparch Spark
• GPGPU accelerated Spatial Operators
• JTS Library
• LocationSpark
• Query Scheduler
• Spatial Partitioning and Scheduling
• Query Executor
• Query Evaluation and Index Cost Estimation
Spatial Big Data : Systems
SpatialSpark : http://simin.me/projects/spatialspark/
LocationSpark : MingJie Tang, Yongyang Yu, Qutaibah M. Malluhi, Mourad Ouzzani, Walid G. Aref:
LocationSpark: A Distributed In‐Memory Data Management System for Big Spatial Data. PVLDB 9(13): 1565‐1568 (2016)
38. 38
• SKSpark : Spatial Keyword Spark by KSNU
• First Spatial Keyword Big Data System
• BRQ : Boolean Spatial Range Query
• 𝑄 𝜓, 𝑟
• 𝜓 = 질의 단어
• 𝑟 = 질의 범위
• 𝐵𝑅𝑄 𝑄. 𝑟 ∩ 𝑜. 𝑟 ∈ 𝐷 ∩ 𝑄. 𝜓 ∩ 𝑜. 𝜓 ∈ 𝐷
• BkQ : Boolean k‐Nearest Queries
• 𝑄 𝜓, 𝑟, 𝑘
• Spatial‐Keyword In‐memory Indexing
Spatial Big Data : Systems
41. 41
• Queries for Static Object/Static Map
• Shortest path query
• Range query : Find the restaurants
• k-NN/top-K queries
• Queries for Moving Object/Static Map
• Continuous range queries
• Location-based k-NN/top-K map queries
• Queries for Moving Objects/Moving Objects
• Internet connected cars
• k-NN/top-K moving objects queries
• Continuous moving objects queries
MO and Spatial Queries
42. 42
Trajectory Queries
• Queries
• Spatial Queries
• Which cars within 1 km from here?
• Spatio‐temporal Queries
• Which cars were within 1 km from here 7am
to 8am today?
• Moving Objects Queries
• Which cars left from the Everland 7am to
8am today?
• Which cars were stayed in Everland more
than 4 hours
• MO Mining Queries
• What cars are moving together for more
than 10 minutes?
46. 46
• Projection Operations
• Spatial Projection
• Temporal Projection
Data Model
LineString sproject( mpoint )
periodtproject( mpoint)
LineString sproject( mpoint )
periodtproject( mpoint)
temporal
projection
spatial projection
t
LineString
start timestamp
end timestamp
47. 47
Trajectory Queries
• Range Queries
• k‐NN Queries
Yu Zheng, Trajectory Data Mining : An Overview, ACM Trans. on Intelligent System and Technology, Sept. 2015
54. 54
Trajectory Queries
• k‐NN Trajectory Queries
• ex ) Car Pool App
• Find similar trajectory cars with my trajectory
Yu Zheng, Trajectory Data Mining : An Overview, ACM Trans. on Intelligent System and Technology, Sept. 2015
55. 55
Moving Objects Indexing
• Indexing Moving Objects
• 3DR‐Tree for Historical Trajectory Data
• Historical trajectories are represented by their three‐dimensional
MBB(Minimum Bounding Box)
Time
Mohamed F. Mokbel, Thanaa M. Ghanem, Walid G. Aref: Spatio‐Temporal Access Methods. IEEE Data Eng. Bull. 26(2): 40‐49 (2003)
Long‐Van Nguyen‐Dinh, Walid G. Aref, Mohamed F. Mokbel: Spatio‐Temporal Access Methods: Part 2 (2003 ‐ 2010). IEEE Data Eng. Bull. 33(2): 46‐55 (2010)
56. 56
Moving Objects Indexing
• Indexing Moving Objects
• Multi‐version Indexing for Historical Trajectories
• Maintain an R‐tree for each time instance
• R‐tree nodes that are not changed across consecutive time instances
are linked together
Timestamp 1Timestamp 0
3D‐R‐tree
Mohamed F. Mokbel, Thanaa M. Ghanem, Walid G. Aref: Spatio‐Temporal Access Methods. IEEE Data Eng. Bull. 26(2): 40‐49 (2003)
Long‐Van Nguyen‐Dinh, Walid G. Aref, Mohamed F. Mokbel: Spatio‐Temporal Access Methods: Part 2 (2003 ‐ 2010). IEEE Data Eng. Bull. 33(2): 46‐55 (2010)
57. 57
Moving Objects Indexing
• TB‐Tree : Trajectory Bundle Tree
• Hybrid index structure which preserves trajectories
• A predominant trajectory index structure for Euclidean
spaces
• Fast access to the trajectory information of moving objects
Dieter Pfoser, Christian S. Jensen, Yannis Theodoridis:
Novel Approaches to the Indexing of Moving Object Trajectories. VLDB 2000: 395‐406
59. 59
Moving Objects Indexing
• The Time‐parameterized R‐tree (TPR‐tree)
• Minimum bounding rectangles (MBR)
• Velocity bounding rectangles (VBR)
• MBR&VBR은 차량의 속도가 유지되는 한 동일한 MBR 안에
유지 될 수 있음을 보장함으로서 update를 최소화
Simonas Saltenis, Christian S. Jensen, Scott T. Leutenegger, Mario A. López:
Indexing the Positions of Continuously Moving Objects. SIGMOD Conference 2000: 331‐342
61. 61
Moving Objects Indexing
• Indexing on Road Network
• FNR‐Tree : Fixed Network R‐tree
• MON‐Tree : Moving Objects on Network
FNR‐Tree : E. Frentzos. Indexing objects moving on fixed networks. SSTD, 2003.
MON‐tree : Victor Teixeira de Almeida, et. al, Indexing the Trajectories of Moving Objects in Networks. GeoInformatica 9(1), 2005
R‐tree
for
Network
R‐tree
for
MO
63. 63
Moving Objects Patterns
• Region of Interest(ROI) : Static ROI
• Fixed pre‐defined regions
• Find frequent visited regions
A
B
C
time
64. 64
Moving Objects Patterns
• Region of Interest(ROI) : Dynamic ROI
• Spatial Clustering and Labeling
• Find frequent visited regions
cluster(x1,y1)
cluster(x2,y2)
time
65. 65
Moving Objects Patterns
• Moving Together Patterns
• Flock
• 특정 Region r내에서 m 개체이상이 같은 방향으로 움직임
• Leadership
• 가장 먼저 움직인 객체
• Convergence
• m 개체 이상이 같은 방향으로
• Encounter
• r내에서 만나는 지역
[37] J. Gudmundsson and M.V. Kreveld. 2006. Computing longest duration flocks in
trajectory data. In Proceedings of the 14th Annual ACM International Symposium on
Advances in Geographic Information Systems. ACM, 35‐42.
[38] J. Gudmundsson, M.V. Kreveld, and B. Speckmann. 2004. Efficient detection of
motion patterns in spatio‐temporal data sets. In Proceedings of the 12th Annual
ACM International Symposium on Advances in Geographic Information Systems.
ACM, 250–257.
66. 66
Moving Objects Patterns
• Moving Together Patterns
• Convoy
• min consecutive timestamps together
H. Jeung, M. Yiu, X. Zhou, C. Jensen, and H. Shen. 2008. Discovery of convoys in
trajectory databases. Proceedings of the VLDB Endowment 1, 1, 1068–1080.
67. 67
Moving Objects Patterns
• Moving Together Patterns
• Swarm
• min timestamps together
Z. Li, B. Ding, J. Han, and R. Kays. 2010. Swarm: Mining relaxed temporal moving
object clusters. Proceedings of the VLDB Endowment 3, 1-2, 723–734.
68. 68
Moving Objects Patterns
• Trajectory Clustering
• A Partition‐and‐Group Framework
J. G. Lee, J. Han, and K. Y. Whang. 2007. Trajectory clustering: A partition‐and‐group framework.
In Proceedings of ACM SIGMOD Conference on Management of Data. ACM, 593‐604
69. 69
Moving Objects Patterns
• Trajectory Clustering
• A Partition‐and‐Group Framework
J. G. Lee, J. Han, and K. Y. Whang. 2007. Trajectory clustering: A partition‐and‐group
framework. In Proceedings of ACM SIGMOD Conference on Management of Data. ACM, 593‐
604
70. 70
Moving Objects Patterns
• Trajectory Class
• TraClass Algorithm
• Classify subtrajectories instead of whole trajectories
Jae‐Gil Lee, Jiawei Han, Xiaolei Li, and Hector Gonzalez, “TraClass: Trajectory Classification Using Hierarchical Region‐Based and
Trajectory‐Based Clustering”, Proc. 2008 Int. Conf. on Very Large Data Base (VLDB'08), Auckland, New Zealand, Aug. 2008
71. 71
Moving Objects Patterns
• Travel Recommendation
• Find the interesting locations and travel sequences from
trajectories
Y. Zheng, L. Zhang, Z. Ma, X. Xie, W.‐Y. Ma. 2011. Recommending friends and locations based on individual location history. ACM
Transaction on the Web 5, 1, 5‐44.
Y. Zheng, L. Zhang, X. Xie, W.‐Y. Ma. 2009. Mining interesting locations and travel sequences from GPS trajectories. In Proceedings of the
18th International Conference on World Wide Web. ACM, 791‐800.
78. GeoVideo : Moving Objects
• GPS Trajectory and GeoVideo의 분석과 모니터링
차량 3대
센서 데이터
비교 분석
79. 79
• MediaQ by USC
• GeoUGV : User‐Generated Mobile Video
• Since 2014
• MediaQ web: http://mediaq.usc.edu
Related Work
Ying Lu, Hien To, Abdullah Alfarrarjeh, Seon Ho Kim, Yifang Yin, Roger Zimmermann, and Cyrus Shahabi,GeoUGV: User‐Generated
Mobile Video Dataset with Fine Granularity Spatial Metadata, In the 7th ACM Multimedia Systems Conference (MMSys), Klagenfurt
am Worthersee, Austria, May 10‐13, 2016
Seon Ho Kim, Ying Lu, Giorgos Constantinou, Cyrus Shahabi, Guanfeng Wang, and Roger Zimmermann,MediaQ: Mobile Multimedia
Management System. In the conference of Multimedia Systems (MMSys), In the conference of Multimedia Systems (MMSys), pages:
224 ‐ 235, Singapore, March 19 ‐21, 2014
80. 80
• GeoVideoIndex by POSTECH
• GeoVideo
• MBTR : Minimum Bounding Tilted Rectangle
Related Work
Dongha Lee, Jinoh Oh, Woong-Kee Loh, Hwanjo Yu:
GeoVideoIndex: Indexing for georeferenced videos. Inf. Sci. 374: 210-223 (2016)
81. 81
• Architecture
PostGeoMedia Architecture
81
PostgreSQL DB
PostGIS PostTrajectory
Trajectory
Type
Trajectory
Indexing
PostGeoPhoto
GeoPhoto
Type
GeoPhoto
Indexing
Geometry
Type
R-Tree on
GiST
BlackBox App for Cars
(SkyBox)
Apache/Tomcat Web Sever
PostGeoVideo
GeoVideo
Type
GeoPhoto
Indexing
Trajectory Storage Trajectory Queries Trajectory Analysis Trajectory Apps
GeoV/P Storage GeoV/P Queries GeoV/P Analysis GeoV/P Apps
Trajectory/GeoPhoto/GeoVideo
Web User
Geospatial Media Contents Management System(GeoCMS)
86. 86
• Temporal Queries
• Query 2 “Retrieve videos in userVideos table which are taken
during ‘period (1100, 1200)
• Query 3 “Retrieve videos in userVideos table which are taken
during ‘period (1100, 1150 )’, and returns the sliced video
files by ‘period (1100, 1200)’ ”
Trajectory Queries
SELECT id, mv
FROM userVideos
WHERE M_Overlaps (mv, 'Period (1100, 1150)');
SELECT id, M_Slice ( mv, 'Period (1100, 1200)')
FROM userVideos
WHERE M_Overlaps (mv, 'Period (1100, 1150)');
87. 87
• Spatial and Spatiotemporal Queries
• Query 4 “Retrieve videos in userVideos table which have entered
the city ‘New York’ ”
• Query 5 “Retrieve the cars in userTrajs table that have left from
city ‘New York’ during ‘period (1100, 1200 )’ ”
Trajectory Queries
SELECT id, name, mv
FROM carTrajs a, city c
WHERE M_Enters (a.mt, c.geo ) = true AND c.name=’New York’;
SELECT id, carnumber
FROM carTrajs a, city c
WHERE M_Leaves(M_Slice(a.mt,'Period(1100, 1200)'), c.geo) = true
AND c.name = ‘ New York’;
88. 88
• Query 5.2 “Retrieve the cars in userTrajs table that traveled
up to 500m from central New York city”
• Similarity Query
• Query 6 “Return LCVS similarity distances between two
videos in userVideos table.”
Trajectory Queries
SELECT id, carnumber
FROM carTrajs a
WHERE M_Max( M_Distance( a.mt, ‘POLYGON(40.74 ‐73.99, 40.74 ‐73.98, 40.73 ‐
73.98, 40.73 ‐73.99, 40.74 ‐73.99 )’ ) < 500.0
SELECT a.id, b.id, M_LCVS (a.mv, b.mv)
FROM userVideos a, userVideos b;
89. 89
• Join Queries
• Query 7: “Retrieve videos whose FOV intersects each other”
• Query 8: “Find the videos which could be taken 'Empire State’
building
SELECT a.id, b.id
FROM userVideos a, userVideos b
WHERE M_ANY(M_Intersects( FOV(a.mv), FOV(b.mv))) ;
SELECT a.id, b.id, a.mv
FROM userVideos a, city c
WHERE M_Intersects( FOV(a.mv), c.geo ) AND c.name =’Empire State’;
108. 108
• Embed in your web document
• Example : http://deepapi.io
GeoCMS : with Deep Learning
Click &
Paste it
<img src="http://api.u‐
gis.net:8002/?command=detect&model=mscoco&class
es=person&img=http://mcalab.kunsan.ac.kr:8080/files
/attach/images/675/386/002/002cfda9daba2f27d0d13
f4e4baf325e.jpg"/>
110. 110
• Road Damage Data Set
• 차량 촬영 영상을 통한 딥러닝 기반 도로 파손 탐지
• 2018 IEEE Big Data Challenge 데이터 기반 딥러닝 : Yolo
• https://bdc2018.mycityreport.net/overview/
Introduction
114. 114
• Deep Model
• Yolo‐based RCNN Model
• Data Set
• Tokyo University, Road Damage Detection Dataset
• 9,053 images, 15,435 bbox
RCNN‐based Detection
Detail Type Class name
Vertical Crack
Wheel mark D00
Construction joint D01
Horizontal Crack
Equal interval D10
Construction joint D11
Alligator Crack partial pavement, overal pavement D20
etc Potholes D40
Paint
Crosswalk Paint Blur D43
LIne Paint Blur(White/yellow) D44
115. 115
도로 파손 지도 생성 알고리즘
mp4
GPS/
Camera FoVs
RCNN 기반
Road Damage Detection
인식된 Road Damage의
location determination
GeoJSON 기반의
Road Damage Map 생성
지도상에 Map의 가시화