Intuvision Event Detection System FORTRECVID 2008

In this paper, we describe the algorithms for three event detection tasks, namely “Elevator No-Entry”, “Opposing Flow”, and “Picture Take” and findings from applying these algorithms to the TRECVID airport data set. We ran a single run for each event detection task. For Elevator No-Entry event detection we used Haar object pedestrian detection followed by histogram matching to find person not entering an elevator. Histogram matching is shown to perform well under pose variations as indicated by our results. In the Opposing Flow event the main challenge was to handle severe occlusions and overlap. In dry run we used Haar based pedestrian detection algorithm but, in the actual run we decided to use background subtraction for detection and tracking. We minimized occlusions and overlaps caused by other objects by considering only the top portion of a door region. Also, heads and shoulder are smaller in dimension so reduce occlusion effects. Restricting the problem in this manner we were able to get good tracking and detected the tracks in the specified direction using our Panoptes “direction event spy”. We incorporated a mechanism to reject outliers in direction event spy to better handle tracking errors which considerably improved the results. Our Picture Take Event algorithm uses characteristic change in illumination caused by flash – large increase in image intensity followed by almost equal drop in intensity for flash detection. Once a flash frame is detected we find the location of the flash and use simple matching algorithm to detect frames where the hands are steady. This approach is not suited for detecting take picture event without a flash, but since in most cases cameras and hands are just barely a few pixels in width and height, it is hard to detect a camera without the presence of flash.