David A. Semko
MSEE, June 2007
Study Goal
The overall goal of the study was to create a system which pre-screens video surveillance feed and assists the user in identifying unusual activity. Towards that goal, we designed an automated scheme capable of performing three large-scale tasks: identifying contacts, tracking contacts, and characterizing contact behavior.
The data used in this study was obtained from the European Community Funded CAVIAR project/IST 2001 37540 [1]. This data was taken from a wide-angled camera in the lobby of the INRIA Laboratories in Grenoble, France. Much of the video contains various pedestrians either moving through the lobby or loitering in certain areas of the lobby. This study attempts to track both types of contacts throughout the scene. A significant event and an alarm is generated to instruct the operator to investigate the area in which the contact was last known to be located when the algorithm is no longer able to obtain a track of said objects.
[1] �CAVIAR: Context Aware Vision using Image-based Active Recognition,� EC Funded CAVIAR project/IST 2001 37540, http://homepages.inf.ed.ac.uk/rbf/CAVIAR/ [website], Last Accessed: June 2007.
Thesis Abstract
A SIMULINK-based algorithm for monitoring contacts in a surveillance video sequence using Optical Flow Analysis and Kalman Filters was developed. The Horn-Schunk Optical Flow Algorithm was used to identify moving contacts in a surveillance video sequence. The position and behavior of these contacts was monitored by a modification of the traditional Kalman Filter. The Kalman Filter algorithm implemented has the ability to track up to ten contacts at a time, correctly assigning each of a maximum ten filters to their respective contacts on a frame-by-frame basis. Initial tests using artificial data show good performance of both the Optical Flow Analysis algorithm and the Kalman Filter Tracking algorithm. Surveillance video data was also used to test the algorithm with promising results.
Video Clips
Artificial data (3 contacts of varying speed and directions), Real data (walk1.mpg from the CAVIAR study, identified tracks represented by green disks, lost contacts identified by a red disk)
For more information, e-mail fargues dot
nps dot edu