Vehicle Recognition Method Based on Color Invariant SIFT Features

摘要

Both aiming at the problems of the color information lack for SIFT features and the interference of color on vehicle identification, a vehicle identification method based on color invariant SIFT features is proposed. Firstly, the color edge regions of the images are calculated by using the RGB information, and the SIFT color invariant feature description vectors are generated in conjunction with the gray information. While ensuring the integrity of the vehicle color features, the method can effectively suppress the false matching caused by the redundant feature points. Then the nearest neighbor matching method is used to find the matching pairs, and the vehicle identification is finished by the discrimination degree weighting strategy. In the vehicle matching experiment with the same type and different colors, the matching number of the algorithm is less affected by the change of the color, and the stability of image matching is enhanced. The vehicle recognition experiment based on 25 kinds of vehicle model database shows that the method can eliminate the interference of the color features on the vehicle identification, and the total recognition accuracy of the method is improved to 88.2%-92.2%.