Summarizing Motion In Video Sequences

1
Summarizing Motion In Video Sequences

• Kevin Forbes
• May 27, 2004

2
Project Goal

• The goal was to summarize the motion of
  foreground objects in video taken by a fixed
  camera against a static background.
• This was done using two major techniques
  pseudo-photographic image processing, and action
  line rendering.

3
Pseudo-photographic Image Processing

• Meant to imitate the work of the photographers
  Marey and Bragaglia.
• Three techniques Multiple Exposure, Long
  Exposure, and Motion History Imaging.

4
Multiple Exposure

• Based on foreground segmentation (threshold
  difference)
• Layer-by-time
• Variable spacing

5
Long Exposure

• Requires segmentation (implemented using
  alpha-blitting)
• Layer-by-time (again)
• Variable exposure time, falloff

6
Multiple Long Exposures
7
Motion History Images
As seen in J. Davis and A. Bobick. The
representation and recognition of action using
temporal templates, 1997.
8
Action Line Rendering

• Used in comics and technical illustration
• Formulate as a search problem

9
Line Model

• Straight line segments
• Four Parameters Angle, Radius, End Points
• Groups of parallel lines will be favoured by the
  search method

10
Input Images and Line Fitting
11
Line Scoring

• Monotonism
• Timeliness
• Temporal Length
• Spatial Length
• Smoothness

12
Search Strategy

• Candidate Generation
• Regular sampling along angle and radius
• Score all monotonic regions, add positive scores
  to candidate pool
• Candidate Selection
• Sort pool by score
• Add N candidates, starting with highest scoring,
  deferring on intersection

13
Results Good!
14
Results Bad and Ugly!
15
Conclusions

• The pseudo-photographic techniques are limited
  by their reliance upon segmentation.
• Using HDR images, you could more closely emulate
  a camera, but the sequence would need to have
  controlled lighting
• Action line fitting shows promise for
  translational motions, but a more robust line
  model should be used to produce better results
• Possible line models groups of parallel lines,
  articulated line segments, curves, arc segments