Traffic Surveillance Video Based Road Condition Analysis

With the development of society,people pay more attention to intelligent transportation.The state analysis of traffic surveillance video becomes more and more important.To extract traffic parameters,the methods of vehicle speed monitoring can be divided into induction coil loop speed detection,laser speed measurement,radar speed measurement and methods by surveillance video.For the traffic state analysis,most of the existing researches are achieved by calculating the traffic parameters firstly and then analyzing the traffic state.Compared with the induction coil loop analysis method which needs more hardware facilities,traffic state analysis method based on surveillance video has the advantages of low cost and high efficiency.As the traffic network is now well developed,and traffic cameras are throughout the road all over the corner,so the number of traffic surveillance videos is very large.These videos should be cleaned regularly due to the large number and the huge space occupied.Coupled with the shortcomings of existing traffic video analysis method in low efficiency,low accuracy and poor applicability,resulting that a large number of the traffic surveillance videos are not been effectively analyzed and utilized.A road condition analysis method based on traffic video is surveillance proposed in order to improve the utilization efficiency,accuracy and increase the practicality of traffic monitoring video analysis.Traffic parameter extraction is divided into three modules: vehicle detection,vehicle tracking and vehicle speed measurement.A method of traffic parameter extraction based on motion segmentation is proposed.For traffic state classification,the traffic state is obtained by using the learning method directly,and a traffic state analysis method based on Grassmann manifold is proposed.The main work and innovation of this paper are shown as follows:(1)We use particle group segmentation(or merger)method in vehicle tracing and we integrate and improve the vehicle detection,vehicle tracking,and vehicle speed calculation methods.The proposed method is superior to the traditional traffic parameter extraction methods by the experimental results of Beijing airport expressway.(2)According to the spatial and temporal character of traffic video data,the Grassmann manifold and neural network are combined.As the traffic surveillance video data has strong temporal and spatial correlations,and neural networks can be a good representation of these correlations,so the combination of Grassmann manifold and neural network can reduce the computational complexity and improve the accuracy of classification by the spatial and temporal correlation of traffic video data.(3)According to the methods of vehicle detection,vehicle tracking and vehicle speed calculation,we build a real-time intelligent status analysis system based on traffic video.The system can analyze traffic surveillance video in real-time,then it can calculate the number of vehicles,the average vehicle speed,occupancy and other traffic parameters.Experiments show that the traffic parameter extraction method based on motion segmentation and the traffic state analysis method based on Grassmann manifold have good performance,and it has strong innovation and practical value.At the same time,we provide a new way of thinking for traffic video analysis.