Research On Vehicle Detection And Tracking From Video Sequences

Vehicle detection and tracking is the basic of traffic management, nowadays, there is no standard method developed. Although many researches have been done on the topic, no satisfy result has been found yet. Our focus in this thesis is the detection, tracking and velocity estimation of vehicles from video sequences.Previous and widely used approaches in motion detection and tracking are: the color and pattern based method, template matching, blob tracking, contour tracking, background subtraction, Gaussian mixture model and optical flow. But those approaches present disadvantages either in the efficiency or in the implementation cost.In this thesis, we used background subtraction technique combined with a morphological image reconstruction and regions of interest filtering for vehicle detection. The used of the regions of interest filtering and the image morphological reconstruction aim to correct error of the background subtraction in object detections. For the vehicle tracking, we use object bounding features to trace vehicle and the centroid method for vehicle velocities estimations.The experimental results of the above mentioned approach prove that our algorithm present both satisfaction in the efficiency and implementation cost. It also shows a great potential to be implemented in traffic surveillance system.