WP1 Apron Activity Model Overview

1
Data Set and Ground Truth
2
Data set and Ground truth

• Data set contents
• Data set resources
• Ground truth contents
• Ground truth resources

3
Data set contents

• Goal
• Capture realistic video sequences with graduate
  difficulties
• lighting variations,
• occlusions
• covering predefined scenarios.

4
Data set contents

• Video providers, main characteristics
• Inrets Building Entrance, Indoor/Outdoor,
  Multi-view
• CEA Corridor Road, Indoor/Outdoor,
  Visible/IR
• RATP Metro, Indoor, Crowded environment
• iLids Road, Outdoor, Real street context
• Silogic Aircraft parking zone, Multi-view

5
Data set contents

• 5 Topics
• Apron
• Building Corridor
• Building Entrance
• Metro
• Road

6
Data set contents
2 points of view
APRON
Different weather conditions
7
Data set contents
BUILDING CORRIDOR
Visible and IR clip
8
Data set contents
BUILDING ENTRANCE
Entrance and Car Park
4 points of view
9
Data set contents
METRO
Real subway scene
10
Data set contents
ROAD
Visible and IR clip
Real scene
11
Data set and Ground truth

• Data set contents
• Data set resources
• Ground truth contents
• Ground truth resources

12
Data set resources
Data set 1 2 Total
Nb sequences 5 AP 3 BC 19 3 BE 4 MO 4 RD 5 AP 3 BC 21 3 BE 5 MO 5 RD 10 AP 6 BC 40 6 BE 9 MO 9 RD
Nb clips (x Nb camera) 14 AP 6 BC 41 9 BE 4 MO 8 RD 10 AP 6 BC 37 9 BE 5 MO 7 RD 24 AP 12 BC 78 18 BE 9 MO 15 RD
13
Data set resources
Data set 1 2 Total
Total sequences Duration 38 min 03 s 12 min 04s AP 06 min 40s BC 02 min 24s BE 11 min 52s MO 05 min 03s RD 36 min 58 s 06 min 33s AP 07 min 24s BC 02 min 46s BE 10 min 16s MO 09 min 59s RD 1 h 15 min 01 s 18 min 37s AP 14 min 04s BC 05 min 10s BE 22 min 08s MO 15 min 02s RD
Total Duration 1 h 23 min 19 s 1 h 00 min 02 s 2 h 23 min 21 s
Mean Duration / sequence 1 min 56 s 02 min 25s AP 02 min 13s BC 00 min 48s BE 02 min 58s MO 01 min 16s RD 1 min 45 s 01 min 19s AP 02 min 28s BC 00 min 55s BE 02 min 03s MO 02 min 00s RD 1 min 50 s
14
Data set resources
Data set 1 Participation (13 participants) 2 Participation (16 participants)
Priority sequences 1 AP 0 BC 1 BE 1 MO 1 RD 92 --- 69 69 100 1 AP 0 BC 1 BE 1 MO 2 RD 94 --- 94 75 94
Wished sequences 1 AP 1 BC 1 BE 2 MO 0 RD 46 15 8 0 --- 1 AP 1 BC 1 BE 2 MO 1 RD 25 25 19 22 19
Others AP BC BE MO RD 25 0 7 0 5 AP BC BE MO RD 23 15 25 22 19
15
Data set resources

• 2 video data set (test evaluation)
• 22 DVD provided
• Documentation (metrics, annotation rules,
  structures and formats)
• Context and calibration information
• Data base (xls file)

16
Data set resources
Topics Event of interest
AP Vehicle movements empty_area, inside_zone, enters_zone, stopped
BC Person and objects interactions opens/closes (door), picks_up/puts_down, abandoned_baggage, changes_zone, running
BE Person, vehicle and objects interactions (complex scene) opens/closes (door), gets_in/gets_out, go_down/up_stairs, door_control, inside_zone, enters/exist_zone, changes_zone
MO Abandoned baggage in crowded environment Waiting, picks_up/puts_down, abandoned_baggage, people counting, exchange_objects, (ticket machine)
RD Person, vehicle and objects interactions (complex scene) opens/closes (door), gets_(in/on)/gets_(out/of), biking, crossing, walking_with, overtaking
17
Data set and Ground truth

• Data set contents
• Data set resources
• Ground truth contents
• Ground truth resources

18
Ground truth contents

• Context information
• Physical objects of interest and events
  annotation
• Annotation rules provided to participant in a
  document and updated during the project

Examples
19
Data set and Ground truth

• Data set contents
• Data set resources
• Ground truth contents
• Ground truth resources

20
Ground truth resources
Data set 1 2 Total
Nb ground truth files 14 AP 3 BC 34 9 BE 4 MO 4 RD 10 AP 3 BC 32 9 BE 5 MO 5 RD 24 AP 6 BC 66 18 BE 9 MO 9 RD
Sequence time 1 h 11 min 36 s (61 470 frames) 48 min 31 s (56 422 frames) 2 h 00 min 07s (117 892 frames)
21
Conclusion Feedback

• Difficulties encountered
• Definition of event
• Annotation subjectivity
• Proposition for other evaluation project
• Ground truth done before data set distribution
  Easier to identify difficulties that may
  encountered participants and add explanations and
  specific annotation rules
• Ground truth given between the deadline
  submission and the results processing to be able
  to correct ground truth before evaluation
• -gt Time consuming