2015+TRECVID+Workshop: Surveillance+Event+Detection (SED) - - PowerPoint PPT Presentation

2015 trecvid workshop surveillance event detection sed
SMART_READER_LITE
LIVE PREVIEW

2015+TRECVID+Workshop: Surveillance+Event+Detection (SED) - - PowerPoint PPT Presentation

2015+TRECVID+Workshop: Surveillance+Event+Detection (SED) Retrospective+++Interactive+(rSED+iSED)+ Task+Overview National+Institute+of+Standards+and+Technology+(NIST) Martial+Michel,+PhD November+18,+2015 About+the+SED+Evaluation


slide-1
SLIDE 1

2015+TRECVID+Workshop: Surveillance+Event+Detection (SED) Retrospective+++Interactive+(rSED+iSED)+ Task+Overview

National+Institute+of+Standards+and+Technology+(NIST) Martial+Michel,+PhD

November+18,+2015

slide-2
SLIDE 2

TRECVID(2015

About+the+SED+Evaluation

  • Surveillance+Event+Detection+Motivation+

SED+addresses+the+need+for+the+advancement+of+technologies+that+can+ perform+automatic+detection+of+events+in+large+amounts+of+surveillance+ quality+video.

  • Identify+each+detected+event+observation+by:
  • The+temporal)extent)(beginning'and'end'frames)
  • A+decision)score:+a+numeric+score+indicating+how+likely+the+event+
  • bservation+exists+with+more+positive+values+indicating+more+likely+
  • bservations+(normalized)
  • An+actual)decision:+a+Boolean+value+indicating+whether+or+not+the+

event+observation+should+be+counted+for+the+primary+metric+ computation

3

slide-3
SLIDE 3

TRECVID(2015

SED+Tasks

4

  • Requires(application(of(several(Computer(Vision(techniques
  • Involves(subtleties(that(are(readily(understood(by(humans,(difficult(to(

encode(for(machine(learning(approaches

  • Can(be(complicated(due(to(clutter(in(the(environment,(lighting,(camera(

placement,(traffic,(etc.

  • in(a(nonEsegmented(corpus(of(video
  • SED(remains(a(difficult(task(for(humans(and(systems

Retrospective+SED+(rSED): Given+a+textual+description+of+an+

  • bservable'event'of'interest, automatically+detect all+occurrences+of+

the+event+in+a+nonXsegmented+corpus+of+video Interactive+SED+(iSED):+Given+a+textual+description+of+an+observable' event'of'interest, at+test+time+allow+a+searcher+25+minutes+to+filter+ incorrect+event+detections from+the+rSED task

slide-4
SLIDE 4

TRECVID(2015

Evaluation+ Source+Data

1

  • UK+Home+Office+collected+

CCTV+video+from+5+camera+ views+at+a+busy+airport

  • Development+Set:+100+hours+of+

video+/+10+events+annotated+

  • n+100%+of+the+data
  • Evaluation+Set+(SED+’09+’10+’11+

’12+’13):+“iLIDS Multiple+ Camera+Tracking+Scenario+ Training+set” (45h),+ 10+events+ annotated+on+1/3+of+the+data+ (7+events+evaluated)

Controlled Access(Door Waiting(Area Debarkation(Area Elevator(CloseEUp Transit(Area

1 2 2 3 3 4 4 5 5

slide-5
SLIDE 5

TRECVID(2015 6

iLids Dataset+ (45h) Annotated+ Dataset+ (15h)+[used+ until+ SED14]

Annotated+ Dataset+used+ for+SED14+ (3h) UnX annotated+ Dataset+used+ for+SED14+ (8h) EVAL15+(9h,+ 4+cameras,++ 7+events)

UnX annotated+ Dataset+ (30h)

EVAL14+(11h,++ 5+cameras,+++++++ 7+events)

SUB15+(2h,++ 4+cameras,++ 3+events)

slide-6
SLIDE 6

TRECVID(2015

Events+of+Interest

Single+Person+events PersonRuns Someone+runs Pointing Someone points Single+Person+++Object events CellToEar Someone+puts+a+cell+phone+to+his/her+head+

  • r+ear

ObjectPut Someone+drops+or+puts+down+an+object Multiple+People events Embrace Someone+puts+one+or+both+arms+at+least+ part+way+around+another+person PeopleMeet One+or+more+people+walk+up+to+one+or+ more+other+people,+ stop,+and+some+ communication occurs PeopleSplitUp From+two+or+more+people,+ standing,+ sitting,+or+moving+together,+ communicating,+ one+or+more+people+ separate+themselves+and+leave+the+ frame

7

slide-7
SLIDE 7

TRECVID(2015 8 EVAL15 SUB15 iSED rSED iSED rSED

BCMIMSJTU

Shanghai+Jiao+Tong+University+(China)

Embrace PeopleMeet PeopleSplitUp

BUPTMMCPRL (7+years)

Beijing+University+of+Posts+and+ Telecommunications+(China)

Embrace ObjectPut PeopleMeet PeopleSplitUp PersonRuns Pointing Embrace ObjectPut PeopleMeet PeopleSplitUp PersonRuns Pointing

IBM (4+years)

IBM+Thomas+J.+Watson+Research+Center+ (USA)+

Embrace ObjectPut PeopleMeet PeopleSplitUp PersonRuns Pointing CellToEar

IIPWHU

Wuhan+University+X Intelligent+Information+ Processing+(China)

PeopleMeet PeopleSplitUp PersonRuns PeopleMeet PeopleSplitUp

ITIMCERTH

Information+Technologies+Institute,+Centre+ for+Research+and+Technology+Hellas (Greece)

Embrace PeopleMeet PeopleSplitUp PersonRuns Pointing

SeuGraph

Computer+Graphics+Lab+of+Southeast+ University,+Southeast+University+Jiulonghu Campus+(China)

Embrace ObjectPut PeopleMeet PeopleSplitUp PersonRuns Pointing CellToEar Embrace PeopleMeet PeopleSplitUp

TJUMTJUT

Tianjin+University+&+Tianjin+University+of+ Technology+(China)

Embrace ObjectPut PeopleMeet PeopleSplitUp PersonRuns Pointing CellToEar Embrace ObjectPut PeopleMeet PeopleSplitUp PersonRuns Pointing CellToEar Embrace PeopleMeet PeopleSplitUp Embrace PeopleMeet PeopleSplitUp

WARDMCMU (CMU:+8+years)

ITEE,+The+University+of+Queensland+ (Australia) and+Carnegie+Mellon University+ (USA)

Embrace ObjectPut PeopleMeet PeopleSplitUp PersonRuns Pointing CellToEar Embrace PeopleMeet PeopleSplitUp

mcis

Beijing+Institute+of+Technology+MCIS+lab+ (China)

Embrace Pointing CellToEar Embrace

nttfudan

NTT+Fudan (Japan)

ObjectPut Pointing CellToEar

10+Participants:

China: 6,+USA:+2,+Japan:+1,+Australia:+1,+ Greece:+1

4 2 3 3 3 4 2 5 6 6 6 6 6 5 2 1 1 4 5 5

slide-8
SLIDE 8

TRECVID(2015 9

Interpreting DetCurve Results

X:(Rate(FA((Ev/h) Y:(Prob Miss((%) Metrics: 1)(Actual(NDCR( (primary) 2)(Min(NDCR 3)(NDCR@TOER ◼ Sys1: 126(Decisions 32(Correct (94(FA) ◼ Sys2: 3083(Decisions 61(CorDet (3022(FA) ◼ Sys3: 126(Decisions 36(CorDet (90(FA)

slide-9
SLIDE 9

TRECVID(2015

Systems+with+Lowest+Actual+NDCR

EVAL15 SUB15 iSED rSED iSED rSED CellToEar TJUMTJUT WARDMCMU Embrace BUPTMMCPRL BUPTMMCPRL TJUMTJUT TJUMTJUT ObjectPut BUPTMMCPRL BUPTMMCPRL PeopleMeet BUPTMMCPRL WARDMCMU TJUMTJUT WARDMCMU PeopleSplitUp BUPTMMCPRL WARDMCMU TJUMTJUT IIPWHU PersonRuns BUPTMMCPRL WARDMCMU Pointing BUPTMMCPRL BUPTMMCPRL 10

slide-10
SLIDE 10

TRECVID(2015 11

EVAL15 rSED PersonRuns

slide-11
SLIDE 11

TRECVID(2015 12

EVAL15(&(EVAL14(( rSED PersonRuns

! 8(CorDet/(139(FA ! 4(CorDet/(87(FA ! 8(CorDet/(26(FA ! 28(CorDet/(238(FA ! 10(CorDet/(57(FA ! 8(CorDet/(2026(FA

slide-12
SLIDE 12

TRECVID(2015 13

! 36(CorDet/(90(FA ! 61(CorDet/(3022(FA ! 62(CorDet/(552(FA ! 26(CorDet/(44(FA ! 32(CorDet/(94(FA ! 27(CorDet/(100(FA

EVAL15(&(EVAL14 rSED Embrace( (only(repeat(site)

slide-13
SLIDE 13

TRECVID(2015 14

EVAL15(&(SUB15 rSED &(iSED PeopleMeet (only(lowest(DCR( sites)

slide-14
SLIDE 14

TRECVID(2015 15

EVAL15(&(SUB15 rSED &(iSED PeopleSplitUp (only(lowest(DCR( sites)

slide-15
SLIDE 15

TRECVID(2015

Conclusion

  • Improvement+in+correctly+detected+events
  • rSED participation+strong
  • “Group+Dynamic+Subset”+helps+participants+

test+their+algorithms

  • Community+effort+to+update+video+set+for+

quality+and+scale

16