Wavelet transforms for use in motion detection and tracking application

This paper presents, an algorithm using wavelets to detect objects from motion. The algorithm is designed to be part of a vehicle tracking system. The main aim is to integrate information from Gabor and Mallat (1992) wavelet transforms to improve the accuracy and speed of vehicle detection. The Gabor-wavelet analysis allows a rapid estimation of image flow vectors with low spatial resolution. Furthermore, it enables us to produce a histogram over the image flow field, its local maxima providing motion hypotheses. These then serve to increase the accuracy of the subsequent object detection using the Mallat-wavelet transform, which provides the required high resolution. The reliability of the motion detection is improved by integration over time. Since no object model is used, the system can detect even small, disconnected, and openworked objects of arbitrary numbers, such as dot patterns. The algorithm behaves well even in a complex motion situation.