Research On Car Image Retrieval Technology In Traffic Scene

Intelligent transportation system is a hot topic in the traffic field, and automatic car image retrieval is an important research direction in intelligent transportation system. It can be used in police investigation work, to assist accident investigation of escape or other criminal activities in traffic. The research on image retrieval of vehicle in traffic scene has higher theoretical value and application significance. The research of this thesis focuses on the feature extraction and matching technology of vehicle image retrieval.The main research contents of this thesis are as follows:(1)The description and extraction of the SIFT feature from car-face image is studied and the data structure for key points based on SIFT descriptor is established.SIFT has the invariance of image rotation, scaling, translation, affine transformation.To some extent, it can reduce the influence of the scale and the angle between the different target images, owing to the different distance and the angle between the vehicle and the high definition camera in the road.(2)We proposed an image feature matching algorithm of SIFTKeyPre based on the attention degree of car-face image feature key points. According to the different contribution of the base image’s key points to image recognition, the attention degree is given to base image’s key points. We analyze the SIFTKeyPre algorithm of retrieval effectiveness, when it was applied in car-face image sample set. And we also discuss the optimization process of the threshold determined, the convergence value of training strength, and the attention degree impact of key points of license plate on retrieval effectiveness. SIFTKeyPre feature matching algorithm has a better retrieval effect without increasing the time complexity, when is compared with the commonly used Flann and Lowe algorithm, which do not distinguish the effects of different feature points on the matching results.(3)The matching algorithm based on the combination of visual word bag model and support vector machine is applied in car-face image. Firstly, the SIFT feature points are encoded by the visual word bag model, then the K-means clusteringalgorithm is used to express the car-face image of each vehicle as a fixed length vector, and support vector machine is used for matching. Experimental results show that the retrieval effectiveness of BoVW-SVM algorithm is better than Flann and Lowe matching algorithm, and is lower than the SIFTKeyPre algorithm proposed in this thesis. The theoretical anlysis shows that the computation time is the least.