Research On Traffic Density Estimation Method Based On Convolutional Neural Network

In recent years,driving has become the most common mode of travel,the number of cars has increased year by year.And the problem of traffic congestion has become increasengly serious,which has brought many negative impacts on people’s lives and work.Monitoring the dense of vehicles on the road has practical value in alleviating the problem of traffic congestion.Traditional methods for extracting traffic parameters have disadvantages such as difficulty in installation,high cost,and unstable accuracy.The images of traffic monitoring videos are intuitive and the information extraction is simple.Therefore,obtaining vehicle data from the video surveillance images has become a mainstream method for extracting traffic parameters.With the rapid development of computer performance and machine learning algorithms,deep learning has gradually appeared in the vision of scholars,and has become an important algorithm in the field of computer vision.As an outstanding branch of deep learning,convolutional neural network has excellent performance in object detection,image classification,and image segmentation.This paper introduces the convolutional neural network into the traffic density estimation,and improves the performance of the traffic density estimation method in this paper through the optimization of the convolutional neural network model.The main work of this paper is as follows:(1)Aiming at the technical difficulties of artificially designing features and extracting features in traditional density estimation methods,a traffic density estimation method based on a convolutional neural network is proposed.Using a convolutional neural network to take a single frame of traffic video surveillance as input,treat all vehicles on the image as a whole,and output different levels of traffic density according to the density of the vehicles,to achieve the end-to-end density estimation of the whole vehicle from a global perspective.Referring to the existing convolutional neural network models,a convolutional neural network model suitable for the method in this paper is constructed.(2)The convolutional neural network model is optimized based on multi-features and lightweighting to improve the performance of the traffic density estimation method.Firstly,two modules of multi-scale feature extraction and multi-level feature fusion are added to thenetwork model to reduce the effect of camera perspective on the feature extraction of the network model and improve the accuracy of traffic density estimation.Secondly,the bottleneck structure is constructed in the network model,and the small-scale convolution kernel and the global average pooling layer are used,which greatly reduces the parameter amount of the model and reduces the calculation cost of the traffic density estimation method.(3)A traffic image data set TD-Dataset was produced,and using keras framework,the traffic density estimation method proposed in this paper is designed and implemented.It verified the high availability of the improved convolutional neural network in this paper on the traffic density estimation problem,the average accuracy rate can reach 97.1%.