Target Tracking Algorithm Based On Mean Shift And Particle Filter

Video target tracking technology is a hot and difficult problem of research in the fieldof computer vision, digital image processing, machine learning, attracting a growingnumber of researchers. Target tracking technology has been widely used in many fields ofmilitary, security and video surveillance, intelligent transportation, medical diagnostics,and human-computer interaction. With the development of computer technology,computer vision range is growing and will have a huge impact on the production and lifeof human beings.Mean Shift algorithm and particle filter algorithm is used several methods in thevideo tracking, but these algorithms have own shortcomings. When the target movingfaster, adjacent to two goals do not coincide area or target is obscured reappeared, thetraditional Mean Shift algorithm is easy to converge to a local minimum point, can notaccurately track the target, or easily lost target. Found after resampling in the particles, theparticle filter, the large particles of the weight value is copied, the particle positions in thenext frame weights should move in the direction of the target movement, the particle filtertracking the target in a variety of complex situations certain mobile, so spend a value inthe right particles to estimate the position of the target in the next frame, and is bound tobe some error, the proposal distribution of the particle filter and particle search rangetracking robustness than large.Through access to a large number of Chinese as well as English Reference, in-depthand meticulous study and research of the defects of the Mean Shift algorithm and particlefilter algorithm in video target tracking technology, improved the above shortcomings.Research work and innovative research done in this paper is summarized as follows:(1) Details of the current domestic and international video target tracking technology,video target tracking technology commonly used algorithm for the analysis of theadvantages and disadvantages of the various methods of target tracking technologyapplications, and video objects tracking technology in the presence of difficult problemsare summarized.(2) In this paper, the idea of the simulated annealing algorithm into Mean Shift theory,improved Mean Shift algorithm adaptive to iterative heuristic probabilistic search, globaloptimization can be obtained in a greater probability, to overcome the traditional MeanShift The algorithm is easy to converge to a local minimum value of the shortcomings,and video faster moving target or the target is obscured, robustness and trackingperformance.(3) propose a target tracking algorithm based on gray system theory and particlefilter, particle filter output current frame target state estimates, the use of gray predictiontheory, the sequence of states and the current state of the historical target position as inputinformation, predictive value and the current target state estimates, correction theresampling particle states and state transition probability distribution of noise parameters,experiments show, the proposed algorithm has better tracking than the traditionalalgorithm to predict the state of the target on the next frame effect. (4) cross-entropy theory to measure the goals and particles to determine the similaritybetween the model of the characteristics of the area, has a good anti-radiation distortionand anti geometric distortion ability to overcome the coefficient of contraction is moresensitive target and unable to adapt to the target expansion shortcomings, and the graypredictive value, adaptive search range to adjust the particle filter target trackingalgorithm has higher tracking accuracy.(5) the based on Csiszer metric Family and particle filter target tracking algorithm, tostudy and analyze the basic theory of the Csiszer metric family, and Csiszer metric familyof four special circumstances metric programming simulation to achieve, with bettertracking results.