Algorithms Research On Video Moving Object

The technology of video motion tracking has been the focus of research in machinevision. With the development of digital processing technology, the technology has become anew generation of research, has been widely applied to military security, airport security,traffic control and many other areas. In recent years, the moving object position estimates andautomatic tracking research have been paid attention to the domestic and foreign researchersin a complex environment, and many tracking algorithms have been proposed.The object tracking is a complex issue. For example, the sudden change of the objectdirection, the movement of non-rigid object, block the interaction between objects, theinter-block of target object and scene, the angle of the movingcamera, etc., All of this willtake great trouble to tracking process. When confronted with some occasions, it is a verydifficult thing that the calculation of the objects’ precise position of each frame and theanalysis of the moving object. Typically, the process of video tracking is restricted to aparticular environment to deal with.Commonly, the method of object tracking can be roughly divided into two categories:deterministic tracking and random tracking. Deterministic tracking is that the trackingproblems convert to find extreme problem for the sake of the cost function in order to obtainthe optimal solution. The algorithm of Meanshift tracking is belongded to deterministic,which is constantly searching for the area to track the object state. Random tracking includedthe kalman filter tracking algorithm and the particle filter which have became the mainstreamin recent years. The Kalman filter is that predict the state of the object, and then according tothe result of the prediction, estimate the location of state which is a recursive estimationprocess. The Kalam filter system has strict requirements in the posterior probabilitydistribution. Particle filter use the propagation of the random sample of particles in state spaceto approximate the posterior probability that the target state process. Particle filter can beadequate for any non-linear state space model, which the accuracy can be close to optimalestimation. However, particle filter needs to collect a large number of particles in eachsampling-state, resulting in enormous computation, real-time poor and other problems, whichare the difficulties that encountered in practical applications.For the deterministic tracking method, first the theoretical basis of Meanshift algorithm-non-parametric density function estimation method is introduced. And then adetailed description of the realization of specific algorithm process is given, the introductionincludes the target template, candidate template, and other important aspects of the realizationof algorithm. Combined with the actual video tracking process which sometimes the targetobject scale can be changed, This paper describes a solution that change the bandwidth ofthe Gaussian kernel function to solve the problem.For the random tracking, this paper describes two kinds of random tracking algorithms-the Kalman filter and Particle filter. The Kalman tracking, introduces the principle and theory,its mathematical derivation, the reason of Kalman filter’ divergent, unscented Kalman filterand Extended Kalman filter were introduced. For the particle filter, the paper introduces themathematical models of the algorithm and the particle degradation in the sampling process.While describing the algorithm of Kalman filter and Particle filter, this paper describes thetwo algorithms combined Meanshift algorithm respectively. The combination algorithm ofMeanshift and Kalman filter is proposed a thought that divides the target block into severalblocks. The combination algorithm of Meanshift and particle filter is proposed the number ofa particle, the noise variance and the scale of target can change adaptively, and can select thetarget template automatically. Experimental results show that the new algorithm achievesbetter performance when the moving direction is changed or human body is partiallyoccluded.