Abstarct: One of the most important research efforts in Intelligent Transportation Systems (ITS) is the development of systems that automatically monitor the flow of traffic at intersections. Rather than being baxsed on aggregated flow analysis, these systems should provide detailed data about each vehicle, such as its position and speed in time. These systems would be useful in reducing the workload of human operators, in improving our understanding of traffic and in alleviating such dire problems as congestion and collisions that plague the road networks. In this research, a novel approach to vehicle detection and tracking in intersections and urban roads where it is difficult to track individual vehicles in heavy traffic because vehicles either occluded each other or are connected together by shadows. Scenes from traffic monitoring are usually noisy due to whether condition. We use background subtraction baxsed on Gaussian mixture model to extract the foreground. For accurate object detection, an efficient scheme is used to remove shadows and noisy speckles. Because of some difficulties mentioned above, we used a set of training image which have captured automatically, to train a classifier. Decision boundary is computed using Support Vector machines (SVMs). A Histogram of Oriented Gradient (HOG) as a feature descxriptor is used to train an SVM classifier. An occlusion system detects blobs that are suspected of having more than one vehicle in them by morphological analysis. Finally, for tracking the vehicles we used a Kalman filter with a multiple hypothesis tracking algorithm. The proposed system has demonstrated a good performance for intersections and urban roads video sequence with an average percentage of correctly detected vehicles of 76.9 %.