Component-model based detection and recognition of road vehicles

Feature extraction and recognition of a vehicle is always a popular research spot in the intelligent transportation system (ITS) area. Based on this technology, many applications, such as the license plate recognition, brand recognition, driving behavior understanding and so on, can be realized to improve the transportation management-control level. Unlike normal task-oriented vehicle recognition methods, a new feature extraction and recognition framework based on a component-model for vehicles is introduced in this paper, which extracts features from vehicle components with a coarse-to-fine mechanism. This kind of deep learned visual feature can be used for vehicle detection, license plate recognition and brand recognition. Furthermore, a component dataset including 110 different brands of vehicles is built up for evaluation. The proposed method obtained a good performance in the experimental result, which is significant for the practical application.