Results

 Original Video

Extracted Keyframes

The next 12 videos show results from applying the algorithm to a sequence of actions. First the keyframes are selected and then the match is shown using one person from the database. We had perfect recognition,even in the presence of occlusions.

1. Composite Video 1

 Original Video

Extracted Keyframes

 Best Match from Database using Phase Correlation and Viterbi

2. Composite Video 2

 Best Match from Database using Phase Correlation and Viterbi

3. View Change

 Original Video

Extracted Keyframes

 Best Match from Database using Phase Correlation and Viterbi

4. Occlusions

 Original Video

Extracted Keyframes

 Best Match from Database using Phase Correlation and Viterbi

Our technique works on videos like this: demomovie. We are now experimenting with a few known movies.

Computer Vision

Perception and Representation

We are also measuring and modeling human subjects' perceptual representations of characteristics or styles of a movement/action or of its performer -- for a given action "verb" (e.g., walking), the "adverbs" and "adjectives" in a movement grammar.  We are able to obtain consistent measures for even very subtle, complex visual judgments of movement, such as the "attractiveness" of a gait and, further, to quantitatively relate such perceptual dimensions with others, such as for the discrimination of gender from movement.

The Grammars of Human Behavior

PIs : Yiannis Aloimonos & Ken Nakayama      

A project funded by the National Science Foundation (HSD)