Research On Vehicle Tracking Method Based On Video In Complex Road Scene

In recent years,with the rapid development of intelligent transportation system,vehicle tracking technology has been widely used in traffic management,public safety and other fields.Therefore,video-based vehicle tracking technology has become a hot research topic in recent years.However,there are complex changes in moving vehicles,such as vehicle size,attitude changes,as well as in complex road scenes such as intersections,partial occlusion and total occlusion of vehicles,etc.So there are still many difficulties in vehicle tracking technology,how to achieve robust and real-time tracking of moving vehicles in complex road scenes still has important practical significance.Aiming at the technical problems of moving vehicle tracking,the current moving target tracking algorithm is studied in depth,and the target tracking algorithm based on kernel correlation filtering is discussed in detail.On this basis,the advantages and disadvantages of the algorithm are analyzed,and the reasonable improvement is made to adapt to the vehicle tracking in complex road scenes.The main research contents are as follows:1)In the framework of kernel correlation filter tracking algorithm,color features are fused,and classifiers are trained based on histogram of Oriented Gradient(HOG)features and color features respectively to detect targets,and adaptive fusion of features is carried out at decision level to predict the location of target vehicles.2)Based on the improved kernel correlation filtering tracking algorithm,a fast classified scale space tracker is introduced to solve the problem that the kernel correlation filtering algorithm can not continue tracking because of the change of vehicle scale.3)Aiming at the problem of total occlusion of vehicle parts,the occlusion processing module is added.The occlusion detection algorithm is used to determine whether the target area in the current frame is occluded or not.If the target vehicle is not disturbed by occlusion,the tracking method of position and scale filter is continued.If the target vehicle is disturbed by local occlusion,the output of the kernel correlation filtering algorithm will not be credible,so the fusion Kalman filter can predict the position of the target vehicle.4)Relevant filter updates the target template only based on the current frame,and the learning rate parameters can not be adaptively changed.Therefore,an adaptive updating method of learning rate parameters is proposed.The improved tracking method is tested on Visual Tracker Benchmark data set and different road surveillance videos.The experimental results show that the proposed tracking method can achieve stable tracking of target vehicles in complex road scenes.The average tracking error on Carl in Visual Tracker Benchmark data set is only 1.3 pixels.