Research On Vehicle Detection And Recognition Technology Based On Convolutional Neural Network

Recently,China’s annual sales and ownership of vehicles have leaped to the first place in the world.The rapid growth of motor vehicles bring a lot of conveniences to our lives.At the same time,it has also posed a huge challenge to China’s traffic,energy and ecological environment.With the overload operation of urban traffic and the increasing traffic pressure,criminal cases involving vehicles occur frequently,which brings new challenges to public security.In order to ensure public safety,the number of vehicle video surveillance has increased dramatically.However,at present,the monitoring pictures mainly rely on the manual interpretation of all kinds of characteristics of vehicles,which is heavy workload and inefficient.Only relying on road traffic management methods and manpower to solve the problem is far from meeting the demand.In order to solve these problems,this paper proposes a model recognition method based on improved mask R-CNN.First of all,based on the bit vehicle data set,cars data set and some data of Compcars data set,our own data set is expanded and built.The preprocessed image is input into a pre trained neural network to obtain the corresponding feature map.Set the region of interest(ROI)for each point in the feature map to obtain multiple candidate feature regions.Then,these candidate feature areas are sent to the region proposal network(RPN)and the depth residual network(RESNET)for binary classification and BB(bounding box)regression to filter out some candidate feature areas.The rest of the feature areas are processed with roiaiign operation.Finally,these feature areas are classified,regressed by BB and generated by mask.In each feature area,FCN operations are performed and output.In addition,in order to further identify the specific vehicle type,this paper proposes a multi feature vehicle type recognition method which combines the improved yolov3 and the optimized Mask R-CNN algorithm.The local feature data set of vehicle including vehicle logo,lamp,intake gate and vehicle contour is established,which simplifies the network structure of yolov3.At the same time,the generating process of the detection frame and the adjustment rules of the confidence of the overlapping frame in non-maximum suppression(NMS)are improved,which are used for the rough positioning of vehicles.Then the vehicle detection frame generated after positioning is output to the mask R-CNN algorithm,which further optimizes the RPN structure for local feature recognition,and good recognition results are achieved.Finally,this paper establishes a vehicle recognition system based on distributed server,which mainly includes database module,file module,feature extraction and comparison module,message queue module,web module and vehicle detection module.The proposed algorithm is embedded in the system and tested,which verifies the application value of the two methods proposed in this paper.In summary,different vehicle recognition models are established based on convolution neural network.Each model has better robustness,generalization ability and application value.It has a very important practical significance and broad application prospects in improving traffic management,such as illegal inspection,hit and run pursuit,distributed monitoring of suspected vehicles and automatic toll collection.