Research On Vehicle Tracking Algorithm Based On Particle Filter

Intelligent transportation system can integrate various traffic information in real time,accurately and efficiently,improve transportation efficiency and road network through capacity,and play an important role in urban road traffic.And vehicle tracking is an important research topic in intelligent transportation system,vehicle can be obtained through the vehicle tracking information such as speed,direction and movement,thus achieving vehicle accurate and stable tracking,is important for intelligent transportation system.However,there are complex tracking scenarios in vehicle tracking,such as high-speed movement of vehicles,mutual occlusion of vehicles,similar color vehicles and rain and snow weather,which can affect the accuracy of vehicle tracking.Therefore,in order to improve the accuracy,stability and real-time performance of vehicle tracking,this paper makes an in-depth study on vehicle target tracking algorithm.Particle filter is a bayesian inference process based on sequence monte carlo method,it has a significant advantage in dealing with nonlinearity of target state and non-gaussian noise distribution.Therefore,it is widely used in target tracking and becomes one of the research hotspots.Because in the complex tracking scenario,the traditional particle filter algorithm cannot carry out more effective vehicle tracking.So this article on the basis of the further study of the particle filter algorithm principle,for vehicle tracking problem in complex scenes,the basic particle filter tracking algorithm is improved,and the accuracy,stability and real-time performance of vehicle tracking algorithm based on particle filter are further improved.The main research aspects of this paper are as follows:1.Different edge detection algorithms and image denoising algorithms are introduced.At the same time,the edge extraction effect of different edge detection algorithms and the de-noising effect of different de-noising algorithms on video are analyzed.2.The basic principles of particle filter algorithm,mean shift algorithm and Camshift algorithm are described in detail.The tracking accuracy and the running time of the vehicle tracking algorithm based on particle filter and the tracking algorithm based on mean deviation are analyzed.At the same time,the simple experiment analyzed the vehicle tracking algorithm based on Camshift.3.Particle filtering algorithm based on single color feature is not accurate and robust in the complex tracking scenario,a particle filter vehicle tracking algorithm based on median filtering and multi-feature fusion is proposed.Through qualitative analysis and quantitative analysis of experimental results,the accuracy,stability and real-time performance of this algorithm are verified from different aspects.4.Due to the effect of clustering of the mean shift algorithm,convergence in particle samples can be closer to the true location of the target area,so this article through the fusion of mean shift algorithm and the particle filtering algorithm for vehicle target tracking,and on the basis of introducing decision mechanism,namely,by comparing the Bhattacharyya distance with the size of the setting threshold,determine whether to the introduction of the mean shift clustering to particle filter framework,if the judgment results no,by optimizing the particle filter algorithm for vehicle tracking.Select video with different complex degree tracking scenarios,conduct experimental analysis,firstly,the accuracy of vehicle target tracking is analyzed qualitatively,and the stability of vehicle target tracking process is analyzed quantitatively.The effectiveness of this method is verified by the comparison and analysis of the tracking experiment.