Tracking and Classification of Moving Vehicles from a Traffic Video

According to the development of the monitor system, detection and classification are the major areas of interest within the field of computer vision, in recent year. Vehicle detection, tracking, and classification are applied in several applications such as security and surveillance, traffic monitoring, and car parking. However, the performance of the traditional method is adversely affected under certain conditions like to lighting condition, occlusion, and camera position. Recent remarkable developments of deep learning have led to a renewed interest in tracking and classification. In this paper, we evaluate recent research and traditional method into the tracking and classification of vehicles in a traffic video. In the traditional method, we present technologies such as background subtraction, Pixel Connectivity, and Connected Components Labeling to detect and track moving vehicles. Besides, an algorithm was implemented to improve motion speed. After that, the vehicles are classified by shape-based feature. In the next method, a vehicle type tracking and classification system based on deep learning are proposed. Particularly, this system uses You Only Look Once (YOLO) which is a commonly used method for tracking and classification of vehicles from video. The experimental on GRAM-RTM dataset showed the accuracy of methods in the different type of vehicles.