Algorithm Design And Application Research Of Multi-pose Vehicle Recognition

The rapid growth of the economy make people’s living standard have a huge change, people’s demanding for cars is also growing. But the traffic management and environmental problems is accompanied, they have become increasingly severe problems. In many aspects, it is an urgent to find a new way for traffic management, and intelligent transportation system emerge as the times require.In the intelligent transportation system, vehicle recognition has been a hot and difficult topic, there are many domestic and foreign scholars have done certain research on it. This paper is mainly focus on the image of vehicle’s front face whose practical shooting may tilt, summarizes a multi pose vehicle recognition algorithm. According to the algorithm, we design and develop a multi pose vehicle recognition system.The system mainly divides into the vehicle sample feature training process and the recognition of vehicle type. The first part mainly deals with the treatment of training sample set. Getting the position of the license plate comes first, then one hand according to the tilt angle of plate to correct the vehicle image, another hand according to the position information of license plate to intercept the vehicle face region. After obtaining the car face region, extract PCA feature on it and storage it to a local file offline. At the same time the basic information of each vehicle in the sample library is stored into the database. The second part mainly concerned the test sample which will input our system. In accordance with the same method mentioned above to get the vehicle face region, and then extract the feature PCA.After the PCA library of the training samples is loaded, find the best matching five training samples by using the improved k-d tree algorithm. Further extracting the SURF feature of training samples and testing samples in the exhaust gate region, at last recognition results can be obtained by SURF feature matching.This paper use Open CV and VS2010 to build the experimental platform, and take SQLServer2005 as the backstage database. 411 images have been collected to build a vehicle face feature database, also established a table in the database named CARTYPE to store the basic information of these vehicle samples. In the experiment, using 133 pieces of test samples whose vehicle face’s size is 80×30 to test the system recognition rate and recognition time. The vehicle brand recognition rate is 97.74%, and the series recognition rate is 94.74%, the average recognition time is 94.39 ms.The experimental results show that, the algorithm used in this paper is feasible and effective which is based on the combination of PCA and SURF to recognize multi pose vehicle. The recognition rate stand very front compared with the similar algorithm, the system developed by the algorithm has good stability and real-time.