The Research On Vehicle Recognition Based On Geometric Sparse Representation Of Contour

In order to increase the traffic efficiency and improve the traffic situation, intelligent transportation system is becoming the development tendency of future traffic. Accurate detection and correct recognition of passing vehicles on the road are the core functions of intelligence transportation system. The vehicle recognition method based on video has the advantages of wide application range, convenient installation, rich information and so on, so it has an extensive application prospect. Therefore studying vehicle recognition based on video has important theoretical and practical significance. This thesis focuses on vehicle recognition based on video, the vehicle contour feature is selected as the basis of vehicle recognition.The accuracy of vehicle contour extraction will directly affect the subsequent vehicle recognition. To extract accurate contour of vehicle, a vehicle contour extraction algorithm is proposed based on background subtraction, which can eliminate shadow interference and is robust to variation of view angle. Firstly, morphology processing is performed on the binary image of the moving vehicle obtained from background subtraction, then noise in background areas can be eliminated and holes in foreground can be filled. Secondly, this algorithm use ellipse detection algorithm to detect wheels and determine the position of wheels and the parameters of ellipses, then eliminate the shadow interference though removing the contour below the wheels. Finally, the angle of view and the wheelbase can be estimated by the parameters of ellipses, and according to the characteristics of vehicle contour on both sides, the contour can be extracted though the use of geometric transformation. The proposed algorithm can effectively remove the shadow of vehicle, and extract accurate contour of vehicle.The existing vehicle recognition algorithms use contour feature as vehicle type feature, which have high demand on the accuracy of contour extraction and have a low recognition rate. To solve the above problem, a novel vehicle recognition algorithm based on geometric sparse representation of contour is proposed. According to the geometric characteristics of vehicle contour, vehicle contour can be approximated by a trapezoidal and a rectangle, and the approximate contour obtained by the trapezoid and the rectangle has a higher degree of recognition than the original contour. Vehicle contour can be sparse represented as a trapezoidal and a rectangle by using of quantum genetic algorithm. Then the proposed algorithm can extract the geometric parameters of trapezoid and rectangle, and vehicle recognition is finished based on predefined recognition rules. The experimental results show that the proposed algorithm of vehicle recognition is feasible and effective, which has a high recognition rate for many kinds of vehicles and is robust to the accuracy of vehicle contour.Lastly, a simulation system for the vehicle recognition algorithm is designed, and each function module of the system is introduced.