Login
Talks & Seminars
Shape Based Detection and Tracking of People in Videos
Ram Navatia, University of Southern California
Date & Time: April 13, 2006 15:00
Venue: EE Conference Room (1st Floor of EE-Main building)
Abstract:

Detection and tracking of people in videos is an important task for many applications such as surveillance, monitoring and activity recognition. The conventional approach to this task has been to detect motion blobs signified by pixel-wise differences from a learned background model. This approach is effective for motion detection when the camera is stationary, illumination is constant or changes slowly; however, even under these ideal conditions, the motion blobs may not correspond to individual objects which may split into multiple blobs or many objects merge into one. Shape-based analysis can help segment or group blobs into individual objects.

In this talk, we present an alternative approach that performs person detection based primarily based on its shape properties and uses motion only for tracking and some verification. This allows the method to detect stationary objects as well as function in situations with a moving camera or rapidly changing illumination. To handle situations with mutual occlusion, we use a part-based representation for the human shape; thus, people can be detected and tracked even if only some of the limbs are visible.

Detection part of this talk is based on a recently published paper at the ICCV-05; the tracking work will be formally presented at CVPR in June 06.

Speaker Profile:
Ram Nevatia is a Professor of Computer Science and of Electrical Engineering in the Viterbi School of Engineering at the University of Southern California in Los Angeles, California, U.S.A. He also serves as the Director of the Institute of Robotics and Intelligent Systems. He is a Fellow of IEEE and AAAI. Professor Nevatia has been active in computer vision research for over 25 years and made contributions in a number of areas. Currently, he is leading a team of over 20 students, post-docs and research staff in projects on object tracking, event detection and recognition, geo-spatial site modeling and human-computer interaction.
List of Talks

Webmail

Username:
Password:
Faculty CSE IT
Forgot Password
    [+] Sitemap     Feedback