A Short-term Traffic Flow Prediction Method Based On Urban Surveillance Image Analysis

Efficient and accurate traffic flow prediction is the prerequisite for intelligent transportation system.Short-term traffic flow forecasting provides protection for intelligent transportation systems,and traffic information is an important basis for traffic analysis and control.Short-term traffic flow forecasting is based on dynamically acquiring traffic flow time series data to predict future traffic flow status data.The most important part of the intelligent transportation system is the camera monitoring system.In recent years,the number of cameras installed in public places has increased by spurt.Faced with rich and massive camera information,the manual review mode has been unable to cope with the explosive growth of video data.The low application efficiency of massive images has gradually become the bottleneck of image analysis.In order to effectively use the information in urban traffic monitoring images to predict short-term traffic flow,this paper proposes a vehicle detection method based on deep learning and a method for short-term traffic flow prediction.Vehicle traffic monitoring images are respectively detected and counted by vehicles and converted into time series data for short-term traffic flow prediction.Aiming at the occlusion of vehicle targets and the false detection in rainy days,the vehicle detection method based on attention structure is studied,and the monitoring attention module is added to the basic Faster R-CNN to extract the backbone network.The features are significantly enhanced while at the same time making the subsequent structure profitable.The current neural network-based traffic flow prediction method has some disadvantages such as poor adaptability,poor robustness,and inaccurate local feature description of the data due to the embedded part of the manual design features.This paper proposes a short-term traffic flow prediction method based on dimensional weighting for residual long-term and short-term memory networks,which can get rid of the limitations of manual fixed feature structure and realize adaptive modeling analysis of traffic flow data.Experiment proves that the method makes full use of the information of urban traffic monitoring image data.On the Singapore city monitoring dataset,the method of this paper shows better generalization ability and higher accuracy.