Research On Vehicle Recognition Based On The Data Fusion Of Radar And Video

In the complex and changeable traffic environment,the traditional vehicle detection and recognition system based on single sensor is easily affected by many factors,such as bad weather,light changes,and branches shaking,so it can’t meet the needs of the actual traffic scene.Compared with the single sensor system,the multi-sensor data fusion system uses multiple sensors to obtain the multi-directional and multi-character data of the target,which not only increases the feature dimension of the target,but also improves the reliability,information utilization and fault tolerance of the whole system.To avoid the shortcomings of the vehicle detection and recognition system based on single sensor,this paper applies Dempster-Shafer(D-S)evidence theory to the research of vehicle recognition method based on the fusion of millimeter-wave radar and video,and makes an in-depth research on improving the performance of target recognition based on multisensor data fusion technology.The main research work is as follows:(1)D-S evidence theory is introduced for the fusion of millimeter-wave radar and camera.Joussemle distance and discount coefficient method are used to measure and correct the evidence conflict respectively,which will lead to the failure of the fusion results.The acquisition of basic belief assignment(BBA)is the key and difficulty of D-S evidence theory.To deal with this problem,a method of constructing BB A based on fuzzy set theory is proposed.(2)Aiming at the heterogeneous problem of radar data and video data,the process of time alignment and radar data transformation from radar coordinate system to image pixel coordinate system are deduced.(3)Aiming at the problem of vehicle recognition based on the fusion of D-S evidence theory,a method of extracting the number of amplitude unit and the length of range dimension spread from the 15-by-15 interception window is proposed in the selection of radar features.In order to detect the real moving target more accurately and reduce the influence of false alarm and missed detection on subsequent vehicle recognition,a target extraction method based on radar tracking data is proposed.In the selection of image features,a method of extracting the area feature and perimeter feature of the image is proposed.In order to verify the rationality and effectiveness of the above radar features and image features,the performance of the large vehicle and the small vehicle on each feature is analyzed and verified based on the measured data.Aiming at the problem that the radial range of the target affects the area feature and perimeter feature of the image,a range correction method based on radar ranging and data fitting is proposed.(4)According to the BBA acquisition method of D-S evidence theory,feature data set is constructed.According to the data set,the target model and test model of triangular fuzzy number(TFN)are constructed to generate the BBA which is necessary for D-S evidence theory.Through the fusion rules,the data fusion of millimeter-wave radar data and video data is realized.Based on the theoretical analysis and the measured data in this paper,the vehicle recognition results of different algorithms under different scenes are compared and analyzed,and the results show that the proposed algorithm based on the data fusion of millimeter-wave radar data and video has better effect.