Automatic Identification Of Non-motor Vehicle Violations Based On Real-time Video Key Technology Research

In recent years,the central direction of traffic management in our province is to legislate and regulate the management of non-motor vehicles.However,in the field of artificial intelligence and intelligent transportation,the technologies of non-motor vehicle detection,license plate detection and recognition,and automatic identification of non-motor vehicle violations are relatively rare.Compared with the existing technologies of motor vehicle detection and license plate recognition,the technologies of non-motor vehicle detection and license plate recognition are relatively rare.The environment is complex and difficult.In this paper,a non-motor vehicle detection method is proposed.Firstly,the fractional differential method based on Canny operator combined with the difference image is used to extract the moving non-motor vehicle contour in the video,and the extracted non-motor vehicle contour is filtered by skin color filtering detection,geometric feature discrimination and other methods,and the extracted non-motor vehicle contour is used to assist the training.In order to improve the performance of the algorithm,a convolutional neural network is developed.Practice shows that this method improves the detection performance of deep neural network,reduces the residual error rate of non-motor vehicle detection,and overcomes the difficulty of collecting target samples on the basis of maintaining the advantages of deep learning technology in target detection.Based on the non-motor vehicle detection,this paper proposes a non-motor vehicle license plate detection method.In the extracted non-motor vehicle contour map,the non-motor vehicle license plate image is detected by using the methods of non-motor vehicle license plate background color clustering,license plate external rectangle proportion discrimination and so on.This method improves the speed of non-motor vehicle license plate image detection,improves the speed and accuracy of its detection.For the license plate image of the non-motor vehicle license plate based on the above extraction,a recognition method of the non-motor vehicle license plate is proposed.According to the characteristics of the format of the non-motor vehicle license plate,this method uses a boundary extension method of the license plate image digital area image based on the pixel scanning of the sliding window,that is,the horizontal and vertical width of the detected characters is used as the basis to scan the boundary of the surface in turn,and the water is used to The characters that do not enter the digital area image in the square direction are included in the area image,so as to expand the digital area image to the target area image of the non-motor vehicle license plate.Then the boundary expansion method is combined with the Faster R-CNN(Faster Region-Convolutional Neural Networks)network training model,and then the BP(Back Propagation)neural network is used to identify the non-motor vehicle license plate image.The characters and numbers in the target area of.Practice has proved that the recognition method of NML not only uses the frame fitting and classification idea of Faster R-CNN network to filter the interference characters of NML,but also improves the accuracy of BP neural network in the recognition of characters and numbers in NML images,and reduces the difficulty of image annotation before target detection,so as to achieve Ningbo traffic management.The requirements of the Department for the performance of license plate image recognition of non-motor vehicles.Finally,the paper also studies the automatic discrimination technology of non-motor vehicle running red light,and proposes a path extraction method.Based on the adaptive mean shift method and the detection of non-motor vehicle in Chapter 4,the method divides the preset area,calculates the center of gravity pixel of the tracking target,and calculates the distance between the target to be tracked and the angle between the target and the horizontal direction based on the center of gravity pixel.On the basis of collection,we can judge whether there is illegal behavior of running red light for non-motor vehicles,so as to improve the accuracy of automatic discrimination of non-motor vehicle running red light,so as to meet the requirements of Ningbo traffic management department for automatic discrimination of non-motor vehicle running red light.