Research On Optimization Of Automatic Detection System Of Illegal Vehicles Based On Image Processing Technology

As traffic congestion increases,the problem of traffic congestion becomes more and more important.The main sources of accidental traffic jams are accidents and accidental traffic disturbances(for example,traffic accidents,scattered goods,parking,etc.).A number of studies have shown that traffic violations are the main cause of traffic accidents.Therefore,reducing traffic violations could effectively reduce the frequency of traffic accidents.It is particularly urgent to establish an information system for automatic detection and evidence collection of traffic violations.This paper mainly discusses the following three points:(1)Firstly,this paper introduced the moving vehicle recognition algorithm based on feature extraction,then studied the current status and problems of traditional methods in detecting illegal vehicle behaviors.This paper proposed a method based on the feature fusion of directional gradient histogram(HOG)and local binary pattern algorithm(LBP).After that,in order to solve the problem on high and low layer feature fusion,applied principal component analysis algorithm(PCA)to reduce the dimensionality of HOG features.Finally,the classifier support vector machine(SVM)is used to realize the recognition of moving vehicles.The experiment applied enough pictures for training,after the optimization of the experimental parameters,the precision of recognition and the recall rate of Recall reached 96.52% and91.12%,respectively,reached the targets of this experiment.(2)Based on the results of HOG-PCA+LBP cascaded optimization algorithm for moving vehicle recognition,a multi-feature optimized vehicle parking detection and recognition method is proposed,included the intersection of road observation area and vehicle detection position information,detection frame centroid distance information and pixel histogram information,then applied the above three kinds of information to determine whether the vehicle has violations.After the identification of this method,better experimental results are obtained,the accuracy of detection and recognition of illegally parked vehicles is improved,and good practical value has been achieved in actual experimental scenes.(3)Finally,after obtaining the information of illegally parked vehicles,the paper focuses on the detection and recognition of illegally parked vehicle license plates.First is data preprocessing,license plate location and correction,license plate character segmentation,then the license plate character data is recognized by the optimized deep learning Dense Net classification network model.For the image data of haze weather,the dark channel algorithm is applied for defogging processing to improve the clarity of the image.After the experimental implementation of the Keras deep learning framework,the optimized Dense Net network model achieved 99.25% precision in character recognition,which proves the feasibility of the network optimization scheme proposed in the paper and the accuracy of license plate recognition.