Research On Key Technology Of Vehicle Identification With Cross-camera

With the rapid development of national economy,highways,urban roads,the construction of more and more parking,traffic control,security management requirements are increasing on the "video surveillance" smart increasing demands.Single camera on a limited area of the monitoring range,unable to meet the requirements of the people.Therefore,the use of a wide range of multi-camera monitor now become mainstream.In this paper,the camera across the vehicle identification technology,the camera between different vehicle registration,and through experiments to test its effectiveness.As used herein,DPM(Deformable Part Model)vehicle detection method to determine the target vehicle based on test results.This can filter out a lot of background information in the image and reduce the computational complexity,improve the accuracy of vehicle identification.Experimental results show that the method can detect the use of DPM partial occlusion of the vehicle,the vehicle is better than traditional detection methods.In the vehicle search,as used herein,it is based on the bag of words model and Hamming coding method.The main role of Hamming codes are put in the same class characteristic points closer subdivision,to solve the traditional visual bag of words model clustering center selection is too large or too small to match the decline in accuracy caused by the problem,effectively improve the SIFT feature points the accuracy of matching.Also proposed a method for multiple allocations can be reduced to match those of the missing,to further improve the matching accuracy,thereby improving the accuracy of vehicle identification.In a cross-camera vehicle identification,the vehicle converts video frame alignment problem into a network flow problem,the vision and time to be considered as a whole,to reduce dependence on video key frame,making the whole network model is not sensitive to the choice of keyframes.Convert video part alignment problems to a network flow problem,find similar video clips to convert the optimal path for the network flow problem.By combining visual and temporal coherence,consistency makes the system scalable visual and time.Finally,cross-camera vehicle identification proof proposed method can effectively cross-camera recognition accuracy rate of up to 81.71%,the recall rate of 84.1%.This article addresses the cross-section of the camera vehicle identification problems for cross-camera vehicle identification extensive application of the foundation.