The Research Of Vehicle Type Recognition Based On Video Sequence

Vehicle recognition has played an important role in the field of traffic monitoringand Highway Automatic Charging System. Vehicle classification technology based onimage processing, which is informative, wide range of applications and settle the videodetector deficiency and becomes the direction of automatic vehicle classificationdevelopment. A new system based on video sequences and SURF (speed up robustfeatures) feature had been developed in this dissertation.The system advanced in this dissertation can be divided into three parts, whichincludes vehicle detection and video segmentation part, feature extraction and selectionpart and recognition part.In the vehicle detection and video segmentation part, a vehicle detection algorithmwhich is based on combining frame subtraction and background subtraction is proposed.First, two motion regions are detected utilizing three frame subtraction and iterativeaverage background subtraction. Based on the OR operation on the two motion regions,a detection result can be obtained. Furthermore, to eliminate the interference on theresult of non-motion region, connected region is analyzed. According to the connectedregion analysis results, vehicle region can be segmented from the traffic video sequencefinally.In the feature extraction and selection part, SURF algorithm is proposed to presentthe features of the vehicle images. According to the properties of the vehicle images,each image is divided into4blocks from top to bottom, and block SURF mean featuresare extracted from vehicle images. On feature selection part, discrimination capability isanalyzed on the features mentioned above, and then the ten biggest discriminationcapability features are selected in each block for a vehicle image. At last the40dimensions features are obtained, as the classification features of vehicle images.In the recognition part, RBF neural network is used to classify the vehicles. TheZISC78Chips, which are designed by using Zero Instruction Set Computer technology,are applied in the system to realize the RBF neural network. The EZB624PCI neuralnetwork card contained eight ZISC78Chips, is the Hardware experimental platform of the vehicle recognition.In the final part of the dissertation, through the vehicle recognition performanceanalysis, there is a comparison experiment between the classification features which thisdissertation proposed and SIFT feature, and the experimental results show that themethod that the dissertation proposed can better meet the practical requirements aboutrecognition rate and recognition time. This paper has important significance andapplication value to the development of automatic vehicle recognition based on videoimage.