Research On The Video Detection Method Of Violation Behaviors Not Driving In The Guiding Lane

With the rapid development of the Internet industry and social economy,the means of urban transportation has become more open and diverse.Consequently,traffic problems in cities are becoming more and more prominent,as the traffic situation becomes more complicated.Traditional vehicle monitoring facilities have been unable to meet the needs of traffic control in cities.With the thorough development of deep learning methods in the field of computer vision,a variety of vehicle detection and tracking methods have emerged,providing chances to innovate technology for vehicle monitoring in complex traffic scenarios.Focusing on the problem of vehicles not following guidance lanes at intersections,this thesis develops a vehicle monitoring system based on a deep learning algorithm.Compared with traditional solutions,ours is more robust dealing with intricate scenarios,runs faster,uses fewer parameters,and provides higher accuracy.The specific work of this thesis is as follows: The SSD object detection algorithm is used in training regarding vehicles in the road.The residual network is used to optimize the backbone network.Besides,we use cluster algorithm to optimize the scale of the prior box.Data from the DETRAC test set are used to test the algorithm.It is verified that our algorithm has better performance in parameter quantity,running speed and accuracy compared with the SSD algorithm.In order to meet the requirements of real-time performance,we adopt the tracking by detection method in the tracking module.Additionally,we propose new judging conditions to identify the same target in two continuous frames and explain its rationality using statistical methods.At the same time,to avoid detection module failure,we add the multi-scale KCF algorithm to our tracking module,which increases the accuracy at a trivial expense of running time.The tracking module was verified with a test video from five different scenarios.Combined with the vehicle detection and tracking algorithms,the vehicle behavior.