Motion Analysis

1
Motion Analysis Human Figure
Processing video to extract information of objects
3 categories of objectives
Motion tracking
Pose reconstruction
Motion and subject recognition.
2
Motion Tracking
Identification and location of subject
characteristics
Based on features or regions Optical
flow silhouettes
Applications Pedestrian flow traffic flow.
3
Pose Reconstruction
2D or 3D tracking of body parts

• Applications
• Medical - gait analysis
• Sports
• Entertainment

• Usually with geometric model
• DoFs?
• Silhouette
• Color, Textures

• What?
• Entire pose
• Hand arm gestures
• Facial expression lips

With instrumentation MoCap.
4
Recognition
Abstract out the what high level semantics
Extract discriminators Neural Nets PCA
Can be based on tracking or pose reconstruction
What? Gestures - by arm motion Person - by
his/her gait American Sign Language (ASL) - by
finger position.
5
Problems.
Complex, varying environment
Segmentation issues
Occlusion of body parts.
6
Possible simplifying assumptions on environment
Static and/or uniform background
Restriction on movement
Restrict number of objects
Restrict complexity of objects.
7
Segmentation problems
Image quality
Motion blur
Low contrast images
Strong shadows, reflections
Loose fitting clothes - use tight fit, markers.
8
Occlusion problems
3D model, 2D image
Underconstrained problems
Use multiple cameras.
9
Taxonomy
Model based or features Edges Regions colors,
textures
2D v. 3D
Static v. dynamic processing e.g. use nOptical
flow
Single camera v. multi-camera Locate points
uniquely Single underconstrained
10
Processing Variations

• Dynamic update of background
• Changing illumination
• Shadows, brightness
• Update of static fixtures

• Background subtraction
• Contract enhancement
• Shadow removal
• Edge detection
• Region filling
• Clustering
• Silhouette formation

• Body part matching
• Find appendages, joints
• Toe - heel curvature
• Head

Speed Real-time only previous frames Interactive
Batch
Need initialization?
Need previous examples?
Single frame or temporal info.
11
Cambridge Research lab
12
Cambridge Research lab
13
Cambridge Research lab
14
3D Model Optimization
15
3D Model Optimization
Temporal coherence Anatomy Area matching Boundary
matching
Matching features of synthetic image with live
image
16
Gavrila
Multiple cameras Detect edges - remove static ones
17
Gavrila
18
Gavrila
19
MERL
Capture motion snippets Baysian formulation
20
Sminchisescu
21
Sminchisescu
22
Sminchisescu
23
Sminchisescu