| Vehicle detection,tracking and license plate recognition technologies have become important parts of intelligent traffic monitoring system.Vehicle detection technology by video can obtain all aspects of road monitoring information.However,under the conditions of illumination change,target occlusion and weather deterioration,the single video detection can not meet the actual needs,so road vehicle monitoring technologies by multi-sensor fusion emerge as the times require.This also becomes the research hotspot and difficulty of intelligent traffic monitoring system.In this paper,we focus on several key issues,such as multi-target vehicle detection algorithm based on the fusion of radar detection and video detection,multitarget tracking algorithm based on features fusion,and license plate location and recognition algorithm with low resolution.The main research contents are as follows:To solve the problem of false detection by single sensor in road monitoring scene,a multisource cooperative vehicle detection method based on deep learning is proposed.Firstly,coordinate transformation of millimeter wave radar data is carried out,and then vehicle location and recognition are carried out by video target detection algorithm based on convolutional neural network.For the two kinds of detection results,a multi-target fusion algorithm based on centroid matching is proposed in this paper.The experimental results show that the proposed algorithm can greatly reduce the number of missed detections,false detections and multiple detections detected by a single sensor,and improve the location accuracy of target detection.To solve the problems of false detection and idendity switches in road monitoring scene,a multi-target tracking algorithm based on the fusion of heterogeneous features is proposed.The algorithm adopts the multi-class similarity measure and multi-participation weight feature fusion calculation equation,and designs the setting equation of each characteristic weight.Through the simple prejudgment of the external environment,the state parameters of the environment are set,and the value of the characteristic weighted values are set according to the different parameters.The experimental results show that the proposed algorithm can improve the trajectory interruption,idendity switches and target loss of single sensor video target tracking,and improve the tracking accuracy of multi-target vehicles.To solve the difficulty of vehicle license plate location and recognition caused by the long distance of the target vehicle in the road monitoring scene,a step-by-step extraction algorithm for location and recognition of remote low resolution license plate is proposed.The algorithm adopts shallow convolutional neural network to locate the vehicle license plate roughly,and then uses the boundary localization method to locate the vehicle plate accurately.Finally,endto-end convolutional neural network is used to recognize license plate characters.Experimental results show that the proposed algorithm can locate the remote target license plate and recognize characters accurately.Finally,based on the proposed algorithms of vehicle detection,vehicle tracking and license plate recognition,a complete software system for road vehicle detection,tracking and license plate recognition is designed.The software adopts modularization and multithreading ideas,and different modules can run separately,which is convenient for the subsequent software upgrade and verification of the algorithms of each module.The software system includes vehicle detection module,data fusion module,vehicle tracking module,license plate recognition module and database storage module.At the same time,the software designed a simple and easy-to-operate man-machine interface.In addition,some interfaces are reserved for the secondary development of the software.The design of the whole software system lays a solid foundation for the construction of the follow-up intelligent traffic monitoring system. |
PDF Full Text Request