Research On Moving Object Detection And Tracking In Video And Image Sequence

Nowadays, multiple targets detection and tracking in video image sequence have been the focus research topics of computer vision. The rapid development of computer science, information technology and im age recognition technology enables video monitoring technology can be used in a very wide range of areas such as national defense and national product. T o meet this demand, we studied the moving tar get detection and tracking techniques in this paper.Particle filter algorithm is a kind of Bayesian estimation algorithm. This algorithm has been wildly used in the multi-target tracking area and its advantage is that it can estimate the target state of any model. However, the disadvantage of particle filter is that we need a lar ge number of particles if we want to obtain better tracking results. Accordingly, the calculation will be great. For this situation, the particle filter algorithm has been improved in this paper, and with the improved algorithm we track m ultiple targets in the video image sequences.Firstly, for the circum stances of non linear non-Gaussian tar get tracking, we describe the relevant principles of particle filter, including the principle of non-linear Bayesian forecasting, Monte Carlo theory , particle filter theory, etc. And the particle filtering algorithm is applied to the one-dimensional strongly nonlinear tracking model. Experimental results show that par ticle filtering algorithm can ef fectively track the nonlinear objective.Secondly, an adaptive Gaussian mixture modeling approach has been used to detect moving targets in the video im age sequence. We use a number of Gaussian m odel to model each pixel in the image, and use the EM iterative algorithm to extract and update the background model, in order to separa te out the background and foreground im ages to achieve the detec tion of m oving targets. Accordingly, we apply m orphological filtering to eliminate noise on the differential results, and get satisfactory results.Thirdly, based on the realization of the background modeling and object detection, we use the improved particle filtering algorithm to track multiple targets in video image sequences. The environment that multi-target objectives activate in is m ore complex relative to the single o bjective. The block between targets and lighting changes can cause the deviation of the tracking results. This paper gives the principle of multi-target data association and a joint probability density of multi-objective algorithm, and put this algorithm combined with particle filter. We use the Euclidean distance between Particles which represent different targets to finish the different divisions of the particles.Finally, the algorithm is used in both indoor and outdoor environments, experimental results show th at the algorithm can solve the target occlusion, lighting changes and other issues, and we achieve the correct tracking of multiple targets.