The Research Of Video Vehicle Tracking Algorithm Based On Compression Sensing

Video vehicle tracking is an important link in the research and development of intelligent transportation system.At present,the research of video vehicle tracking algorithm has been advanced,and a lot of research results have been obtained,but there are still a series of problems.Because of the complexity of real traffic scene,a series of scene interference problems are often encountered in the video vehicle tracking process,shadow and target interference are the most common problems that will give rise to tracking failure sometimes.For these two types of problems,we design a video vehicle tracking algorithm based on compression perception which framework is real-time compression tracking algorithm and the real-time and high efficiency is also considered.The research work mainly includes five aspects.(1)The relevant theoretical knowledge.Collecting many Chinese and English literatures and well understanding of video vehicle tracking principle and related theory of compression tracking for the careful analysis of the commonly used video vehicle tracking algorithm at present.(2)Sparse measurement matrix analysis.This matrix is used to reduce the high-dimensional original signal to obtain the low-dimensional observation vector.The minimum loss of original signal should be guaranteed in the process of dimensionality reduction.The high-dimensional haar-like feature of the target vehicle was reduced by using the random Gaussian matrix,and the features of the reduced dimension were used to update and classify the classifier.(3)Real-time compression tracking algorithm.Real-time compression tracking is a real-time tracking algorithm for moving target,which has the advantages of high tracking speed and low computational complexity.When using this algorithm achieves target tracking,first of all,there need sampling in the process of tracking,then the target feature extraction,using the classifier to classify the sample processing,has the largest classification parameter of samples is the target location.However,the tracking effect is not ideal when the algorithm meets the problem of shadow and target interference,so it need to be improved.(4)The design for target location optimization algorithm.Aiming at the more common shadow and target interference problems encountered in the video vehicle tracking process,the target position obtained after classifier classification was optimized.The target coarse area are divided into subregions for recalculating the regional center of mass to obtain the target location by a series of judgment,and the problem of target interference is deal with through the weighted sequence tracking window.In addition,the weighted sequence tracking window corrects the error of the center of mass calculation and improves the accuracy of the algorithm.(5)Experiment for algorithm verification experiment.Matlab R2014 a development platform and other auxiliary software are used to verify the feasibility and accuracy of this algorithm.The accuracy of the algorithm is verified by drawing the curve of the distance change curve and the coordinate position of the center.By calculating the success rate,the efficiency of the algorithm is verified.The experiment proves the feasibility and effectiveness of this algorithm.There are two research innovation:(1)For the shadow problems,the mode selection and the center weight calculation are added on the basis of the original algorithm,that is dividing the location into several subdomain after determining the approximate location of the target and the molecular region is retained to recalculate the center of mass by a series of conditional judgment,to obtain goal position and improve the robustness of the this algorithm.(2)When using the weighted sequence tracking window improve target interference problems,the continuous multi-frame tracking window position that contain target are taken for weighted processing.On the one hand,the problem of target interference is been dealt with,on the other hand the error in the process of calculating the middle quality of the shadow processing is corrected that improve the accuracy of the algorithm.The deficiency of the algorithm design is that the tracking will be failure if the target is lost,classifier will be updated according to the wrong target characteristics.In addition,if the video image sequence exists large area coverage for target vehicle,the algorithm cannot achieve efficient tracking of the target.