Vehicle Component Detection Based On Deep Learning

With the development of social science and technology,intelligent transportation system plays a more and more important role in traffic violations and traffic accidents.Among which the vehicle search system is an important part.The detection of vehicle components is to detect the position of each part of the vehicle,which has a decisive impact on the entire vehicle search system.From this point of view,this paper proposes a vehicle components detection method based on Deep learning Model,which preserves the position relationship between components,and effectively improves the validity and reliability of the system.In view of the breakthrough of the deep network model in image processing,speech recognition,search engine field and successfully applied in the field of target detection,this paper uses deep network model to detect vehicle parts.Firstly,this paper will regard vehicle parts as vehicle key points,and for the first time applly three cascaded convolutional neural network(DCNN)to realize the key point detection,based on this,the Faster RCNN vehicle Parts detection method with the constraint of the center points of the vehicle parts is proposed.In this method,DCNN is used to realize the location of the key points,and then the candidate region filtering mechanism and output strategy of the Faster RCNN are optimized by the positioning result.Therefore,the quality of the candidate areas are greatly improved,and the performance of the vehicle components based on Faster RCNN is improved obviously.Then,on this basis,this paper continues to improve the detction network,firstly,we use first-level convolution network of DCNN to generate vehicle key points,and then use multi-scale and dense sampling strategy to generate effective candidate regions in order to replace the Region Proposal Network of Faster RCNN,and a Fast RCNN vehicle component detection method with vehicle component key points generating the candidate region is proposed.This method can effectively reduce the network complexity,improve the training efficiency,and improve the detection performance of vehicle components.In this paper,the improved deep network model is used to effectively improve the detection performance of vehicle parts of the traffic data set in the existing real scene,while reducing the time of single frame detection,the reliability of vehicle search is also effectively guaranteed.