Highway Congestion Identification Based On AlexNet Model

With the rapid development of economy and the increase of car ownership,the congestion of expressway is more and more frequent,traffic delay caused by traffic jam is more and more serious,so it is urgent to solve traffic jam.The congestion detection method based on image video can not destroy the road,install the traffic without interruption,and the detection feature is comprehensive and able to record the scene image.In the past,the method of expressway congestion recognition based on video analysis has the disadvantages of high latency,slow recognition speed and high recognition cost.This paper uses the video frame(image)analysis to achieve traffic congestion status identification,can effectively avoid the above shortcomings.In recent years,convolution neural network in image recognition has shown the advantages of good versatility,high recognition accuracy,wide application range and so on.In this paper,a method based on convolution neural network alexnet model and image for expressway congestion identification is presented,and the main research contents are as follows.(1)A general congestion image database of highway is established.In the paper,some video images of Xi ' an and Hangzhou's highway are taken as a sample of congestion image of highway,and the samples are labeled as congestion and unblocked.The database contains more than 46,500 samples and more than 800 scenes,which contains the congestion status of 23,000 samples and 300 scenes,and it include 23,500 samples of smooth state and 500 scenes.the database also contains the image in the situation of rain,snow,fog and other weather,and the morning,noon and dusk,(2)Research on high speed road congestion status identification based on Alexnet.Firstly,the AlexNet model is trained by the congestion database of highway,and the recognition accuracy and convergence of AlexNet model are tested and evaluated by adjusting learning rate,model layer,and volume kernel size.The experiment shows that when the initial learning rate of the model is 0.01,the model layer is 8 layer,the first layer volume kernel is 11x11,the accuracy of the test sample is 96.5%,and the test time of the single sample is 0.03 s.(3)The application of the road congestion identification method based on AlexNet in the real highway.Firstly,we established the real-time image transmission between the servers by using.NET Framework and Network transport protocol,the initial test accuracy of the actual data is 90.1% by using the AlexNet model.Because of the change of road background color and weather in the actual situation,this paper adds it to the original database to adjust the model,and the accuracy of the adjusted model in the actual test is 96.4%.The traffic state recognition method based on AlexNet model can effectively identify the congestion state of traffic scenes,and it can avoid the manual design features of specific scenes,and it is efficient and convenient to use in practice.