Abstract

In this paper, we propose un automated vision-based traffic surveillance system. The system works on one-direction roads for vehicle counting and moving- contrariwise vehicle detection. Objects are extracted by using a hybrid method of background subtraction and temporal differencing, which consider each object as a connected component. The proposed multiple-object tracking method finds object correspondences by template matching. Our surveillance system works in real time at approximately 12Hz for 352x288 resolution color video on an Athlon XP 2000+ PC