Algorithm And System Implementation Of Vehicle Recognition On Video Sequence Based On Deep Learning

Vehicle recognition,including vehicle classification,brand fine-grained recognition,which plays an important role in "black car inspection","big cars and small signs" that strike toll,and tracing the escape vehicle.In traffic monitoring video,the accuracy of the traditional recognition detection algorithm is not high and the robustness is not strong because of the changes of external illumination,the motion target blurred,the vehicle engine cover reflecting.To this end,the paper,relying on the "the system and application of toll lane vehicle characteristics recognition authentication based on deep learning" provincial project,has launched the application research of vehicle recognition algorithm,which has important application value and theoretical significanceThe paper describes the status of vehicle classification detection and car brand recognition at home and abroad,analyzes and compares the performance of several traditional target detection algorithms,and points out the advantages and disadvantages of these algorithms.On the basis of analyzing the mechanism of deep learning network model and combining the application scene of vehicle identification and detection,the paper gives a vehicle detection optimization algorithm based on the deep learning network model YOLO.The actual test of Expressway or ETC toll shows that the accuracy of vehicle classification detection is obviously improved,and the real-time requirement of detection is satisfiedThe paper analyzes the limitations of the important region of the location image and the refinement of the fine-grained image feature expression method,and then studies the network model algorithm of RA-CNN.The algorithm uses attentional extraction network APN to extract the most discriminative discriminant regions,so as to further extract the subtle differences among different brands of vehicles.The combined algorithm improves the accuracy of fine-grained recognition.The model is trained and tested on the Stanford Cars-196 dataset.It shows that the precision rate of the fine-grained recognition of the vehicle is nearly 90%.Considering that the vehicle classification in the Cars-196 library has not included all the model brands,and the classification accuracy of the ETC scene is low,the paper divides the vehicle brand into 73 categories with the CompCars data and the category of the road traffic vehicle,and eliminates the information part of the non vehicle in the data set picture.In addition,1500 pictures of surveillance vehicles,including light and dark changes,are added to the actual scene,and the categories are marked separately.The paper improves the structure of RA-CNN network and trains the network model through two recursive convolution feature extraction.Measured in the ETC scenario,the accurate recognition rate of vehicle brand is 95%.The paper integrates the two algorithm module of vehicle classification detection and positioning and the fine-grained recognition of vehicle brand to form a complete vehicle video recognition system:the vehicle classification detection and positioning module is used to detect the picture.Once the detected vehicle belongs to the Car category,the vehicle local picture is completed by the fine-grained recognition network to complete the vehicle.Fine-grained recognition of the brand,and then realize the function of the whole vehicle recognition system.The innovative points and contributions of the paper are as follows:?It optimizes the vehicle detection algorithm of deep learning network YOLO model,and improves the accuracy and real-time performance of multi scene vehicle video recognition.?The network model of RA-CNN is optimized to improve the processing speed of vehicle brand recognition and the recognition rate of vehicle classification.?Combined YOLO vehicle classification detection algorithm and car brand fine-grained recognition algorithm,to realize the vehicle video recognition and detection at the same time,for the car model to fine-grained recognition of the vehicle brand.