Real-time implementation of a vision based automatic push button

Abstract

This thesis presents a real-time digital signal Processing (DSP) system to detect the presence of pedestrians waiting to cross an intersection. Such a system would improve both pedestrian and vehicle flow considering that almost one third of people do not press the push button and there is no need to activate a walk time if no pedestrian is present. This system employs videoimaging as its sensing mechanism since many sites have already been equipped with videocameras and hence no additional hardware installation is required for its deployment. A pedestrian detection algorithm is developed and implemented on a TMS32OC40 DSP chip such that videosignals captured from an actual intersection are processed in real-time. Basically, the algorithm consists of a background update and a background subtraction procedure applied to image blocks. The background update is done in order to cope with brightness changes that occur in actual outdoor scenes. In addition, waiting and moving pedestrians are distinguished by using the temporal intensity changes of image blocks. The DSP implementation involves full utilization of the available resources of the DSP architecture to minimize the processing time. This implementation has led to a detection accuracy of more than 90% at a processing rate of 3 frames per second.