February 12: Chris Bregler, CIMS
Reading Motion Signatures
This talk will cover recent research in "human motion style" analysis
based on new computer vision, motion capture, and statistical models.
There is anecdotal evidence that people can determine someone's
identity and attributes based soley on motion patterns. These motion
signature styles are known as an "intrinsic biometric." We have
developed several computational solutions based on high-resolution and
low-resolution camera input, and a new statistical "motion signature"
model. Most recently, we also demonstrated how to significantly
improve other hard-biometrics (such as face and voice recognition)
with these new motion modalities.
This talk includes joint work with George Williams and Graham Taylor.