Tree-based vehicle classification system

Nowadays, traffic surveillance systems are installed in major cities. They are usually used for two purposes, i.e. realtime traffic monitoring and archived events searching. For the latter purpose, the traffic surveillance systems can be used for police officers' benefits, such as vehicle identification in specific events including stolen vehicles or hit-and-run cases. In such circumstances, the officers are required to identify the vehicle in archived videos according to its appearances. This task is usually accomplished manually through visual perception. The problems arise from this approach Even though this approach results in good accuracy, it is time consuming and prone to error due to human fatigue for long duration videos. In order to solve these problems, a tree based vehicle classification system is proposed. This system consists of three modules, i.e. feature extraction, classification, and search manager. The feature extraction module is used for image and video processing. It extracts the desired features to be used further in the classification module. The classification module uses these features and results in pre-defined vehicle classes. The classification results are stored in the search manager module for further filtering according to user's query command. This paper focuses on the classification module. There are two features designed to be used in the proposed classification module, i.e. types and colors. Vehicles are classified into four classes of type and seven classes of color. Several tree based algorithms are applied to the dataset. The experimental results show that all the algorithms are comparable. However, the highest accuracy for type and color classification are obtained by using decision tree and bagged decision tree, respectively.