Detecting and Numerating Vehicles from CCTV Traffic Camera Movies Using a Support Vector Machine

Nowadays closed circuit televisions (CCTV) have been highly developed and have been utilized in most of road intersections and places with heavy congestion. CCTVs are very useful for automation the traffic control. By vehicle detection and computing the number of vehicles automatically from CCTVs; controlling urban roads, as well as accident management with high accuracy and speed would be possible. In this paper, a different procedure for detecting and numerating vehicles from CCTV traffic camera movies was proposed. In this regard, at first a region of roads was detected using the frames of a short part of CCTV movies. Then in order to gather the training data in the detected region, some features were selected according to their ability in clarifying vehicles. Afterwards, a support vector machine (SVM) and an artificial neural network (ANN) were proposed for detecting vehicles. Finally, the number of vehicles were computed by binary results of detected vehicles. Comparing the results of the proposed ANN-based method with the proposed SVM one reveals that the proposed SVM-based method has a better performance in computing the number of vehicles in cameras that have a long distant from vehicles.

Vehicle detection, Road region, Support Vector Machine, Artificial Neural Network, CCTV camera movies