Skip to main content

View-Invariant Representations for Human Activity Recognition

The representation plays an important role in recognition and understanding of human action from video sequences. A view-invariant representation of action consisting of dynamic instants and intervals, which is computed using spatiotemporal curvature of a trajectory, is presented. In order to validate our representation, we report experiments on several different actions performed by different people, and captured in different viewpoints.

Result Videos

 

Video Synchronization
Input #1: wide base line videos
Synchronized video
Input#2: moving camera
Synchronized video
Dance Sequences — Input unsynchronized videos
The synchronized video: All the sequences are warped towards the upper right one
The synchronized video: All the sequences are warped towards the lower right one

 

Related Publications

Cen Rao, Mubarak Shah, and Tanveer Syeda-Mahmood, Action Rectionition based onView Invariant Spatio-temporal Analysis, ACM Multimedia 2003, Nov 2-8, Berkeley, CA USA.

View-invariant Alignment and matching of Video Sequences, The Ninth IEEE International Conference on Computer Vision, ICCV 2003, Nice, France

Cen Rao, Alper Yilmaz, and Mubarak Shah, View-Invariant Representation And Recognition of Actions, International Journal of Computer Vision, Vol. 50, Issue 2, 2002

Cen Rao and Mubarak Shah, View Invariance in Action Recognition, Computer Vision and Pattern Recognition, CVPR 2001, Kauai, Hawaii, Dec 11-13, 2001

Cen Rao and Mubarak Shah, View-Invariant Representation and Learning of Human Action, IEEE Workshop on Detection and Recognition of Events in Video, Vancouver, Canada, July 8, 2001

Cen Rao and Mubarak Shah, A View-Invariant Representation of Human Action, International Conference on Control, Automation, Robotics and Vision, Singapore, Dec 5th-8th, 2000

Power point presentation is here

For more information, click here