Typical Traffic Vehicle Classification Based On Sparse Recognition Of Class Dictionary

With the fast development of social economy and the improvement of people’s living standard, the transportation discharge increases continuously and a variety of traffic problems occur frequently. Therefore, the Intelligent Transportation System is particularly important.As the carrier of the whole traffic, the traffic vehicle is the key point of the intelligent transportation system.In order to realize the effective management of the traffic vehicle,the scientific and reasonable intelligent classification of typical traffic vehicle is one of the research hotspots in the fields of information, communication, automation, computer and so on.Based on image feature extraction, sparse encoding and pattern recognition theory, this paper proposes a new classification system based on the sparse recognition of class dictionary. The system is used to classify the three typical traffic vehicles, such as bicycles, motorcycles and cars. Sparse recognition of the global dictionary is constructed by extracting the feature of the training samples. Using global dictionary contained in SURF feature of the training samples to sparse recognize.A classification system based on sparse recognition of class dictionary is proposed to improve the system indicated above. In the aspect of feature extraction, SURF algorithm based on FCM clustering is proposed, which makes all feature vectors have a fixed dimension.In sparse recognition, there is a large amount of redundancy in the global dictionary information, which is not very good to highlight the local characteristics of the traffic. In order to improve the recognition rate, a sparse recognition method is proposed, which introduces the class dictionary and the dictionary learning, in order to better express the local characteristics of the vehicle.The system was verified by three groups of contrast experiments. Respectively, comparison between the sparse recognition method and the SVM method, global dictionary and the class dictionary, traditional SURF features and improved SURF features.Experiments show that the method proposed in this paper can achieve a better recognition on the three kinds of traffic vehicles, such as bicycles, motorcycles and automobiles.