The Vehicle Video Retrieval Based on the Angle of Discrimination
|School||Central South University|
|Course||Information and Communication Engineering|
|Keywords||The vehicle retrieval Feature Fusion angle discrimination B-SIFT|
The vehicle video retrieval is based on the combination of the video-image processing and pattern recognition technology, which aims to retrieve interested vehicle in the video fast. With the continued development of urban traffic and the surge in traffic capacity, there is a substantial increase in the number of vehicles, followed by various traffic violations, traffic accidents, and other crime related to traffic safety such as vehicle-stealing, smashed and so on. Traditional traffic information collection and retrieval methods have been unable to meet the needs of the development of modern transport. Thus the video-based traffic monitoring system has been developed rapidly. Of the vehicles retrieval methods, most are used for traffic volume statistics, and vehicle identification systems.Vehicle video environment is complex, in which there are various angles of vehicles and non-vehicle movement disturbing objects. This thesis proposes a discrimination retrieval algorithm based on the angle of the vehicle, fulfilling the retrieval of the vehicle in the video image. First use a combination of methods of Gaussian mixture modeling and background difference method to extract key frames; propose a extraction algorithm based on shape feature fusion. The algorithm uses three shape features to identify the target joint discrimination, and adjust the value in accordance with the value of each feature in a retrieval. The experimental results show that the algorithm can correctly distinguish effective frame and interference frame. This thesis puts forward a motion estimation-based vehicle angle detection algorithm. The algorithm uses the area balance method to calculate the motion vector of the vehicle, and analyze the vehicle angle in accordance with the corresponding principle of the motion vector and the vehicle angle. This algorithm can correctly distinguish between the front vehicle and the side of the vehicle.For two different situations of front vehicle and the side of the vehicle, use different search methods for vehicle detection. Use B-SIFT (Block Scale-invariant feature transform) to retrieve the front vehicle. B-SIFT is the improvement of the SIFT algorithm, which aims at the face image for the front of the vehicle to sub-block the face image so that the car face images can match correctly in the same block. This solves the cross-matching problems and improve the correct rate of the front-vehicle retrieve. Combining HSV (Hue Saturation Value) retrieval methods and shape characteristics to determine the side of the vehicle retrieval methods. The algorithm can be retrieved on the side of the vehicle, and has a certain robustness of lighting conditions.