| With the development of the modern high technology and increasing demand of our country's road transportation, the Intelligent Transportation System has become a hot research topic. Automatic Vehicle Identification is an important part of the Intelligent Transportation System. It has wide application in many fields, such as toll gates in bridge and crossing, monitor in parking area, vehicle violation records, and etc. It has tremendous economic value and practical significance.As a key part of Automatic Vehicle Identification, Vehicle Logo Recognition mainly consists of two parts: Vehicle Logo Location and Vehicle Logo Recognition. The existing Vehicle Logo Recognition method focused on the followings: template matching, pixel distribution, edge histogram, edge invariant moment, SIFT features and so on. However, the recognition rate and the recognition speed of the algorithms need to be further improved.In this paper, data dimension reduction methods and sample fusion methods are used in Vehicle Logo Recognition, which is to improve recognition rate and recognition reliability. The main works are listed as follows:Principal Component Analysis based Vehicle Logo Recognition algorithm is proposed in this paper, which is on the basis of comparative data dimension reduction methods. Firstly, Eigen-Vehiclelogos is obtained by using Principal Component Analysis. Secondly, projection coordinates on the Eigen-Vehiclelogos subspace of the known samples are used as the features of Vehicle Logo Recognition. Finally, according to the projection coordinates of the unknown sample, vehicle logo types are recognized by BP Neural network. The experimental results showed that the proposed method can increase the recognition rate.Sample fusion based Vehicle Logo Recognition algorithm is proposed in this paper, which is based on vision psychology. According to the registered vehicle logos of the same class, average vehicle logos are obtained by fused. And then vehicle logo types are recognized by the proposed Principal Component Analysis based method. The experimental results show that the recognition rate is greatly increased compared with the method without using average vehicle logos.The practical experiment results show that the proposed method can increase the recognition rate and reduce the computing cost greatly, and it is robust to noises and illumination variation. |
