Research On Object Detection And Tracking Algorithm Based On Bayesian Framework

Video analysis is one of the hottest and the most challenging issues in the field of moderncomputer vision applications. Its basic purpose is to automatically understand the behavior ofthe monitored area. For most of the video analysis system, the video target detection andtracking is one of the key steps. So the video target detection and tracking algorithm is of greattheoretical and practical significance.Background extraction is a key technology in the video moving target detection. thebackground of the traditional modeling method has the obvious flaws, for example, thebackground difference method is unable to identify the movement information of thebackground, all kinds of mixed Gaussian radial disturbance robustness is poor, the backgroundextraction algorithm based on Bayesian framework is considered to be an effective way tosolve the above problems. this article introduces a kind of moving target detection algorithmbased on Bayesian decision rule, this algorithm is based on the statistical information of eachpixel point on Bayesian background segmentation, at the same time, taking into account thatthe gradient has strong stability characteristics of the light disturbance, this algorithmintroduces the image gradient antijamming module, making the algorithm more robust totransient light.In terms of video tracking, it is no hard to see from Bayesian optimal estimation that,Kalman filtering method is an analytical form of Bayesian estimation under linear Gaussianassumptions, and also the optimal filter for linear Gaussian dynamic systems. Due to Kalmanwith the above characteristics, this paper proposes a continuous Kalman filtering method, agood result of video target tracking can be yielded under occlusion conditions.For target tracking problem of nonlinear non-Gaussian systems, a lot of people putforward various solutions, particle filtering algorithm is a more successful method. This articleproposes a particle filter video target tracking algorithm based on multiple information fusion.In this algorithm, the traditional particle filter algorithm is combined with a variety of featureinformations. This algorithm have excellent performance under complex scenes such as in thecamera is not fixed, large noise, occlusion, etc. The main work in this paper includes:1) Taking into account that the gradient has strong stability characteristics of the lightdisturbance, proposes an anti-transient light disturbance video target detectionalgorithm based on Bayesian framework.2) For the linear Gaussian system, a method based on continuous Kalman filtering isput forward, a good result of video target tracking can be yielded under occlusionconditions.3) For more complex nonlinear non-Gaussian systems, a particle filter video targettracking algorithm based on multiple information fusion is proposed and thisalgorithm have excellent performance under complex scenes.