Multi-vision Cooperative Target Tracking Based On Probability Model

In the field of monitoring,with the continuous improvement of computer hardware,the application of target tracking has become more and more widespread.However,due to the deficiency of the single vision sensor in extracting the target information,the target loss is likely to occur in a scene with confusion or long-term occlusion.Multi-vision collaborative target tracking can effectively solve these problems,especially for the long-term occlusion of the target.Therefore,this paper proposes a multi-vision cooperative target tracking method based on probability model.Through multiple experiments,it achieves better results in the long-term occlusion of the target.The content of this article mainly contains the following three parts:Aiming at the problem of insufficient extraction of target features in single vision,a spatial particle filter target tracking method that fuses the posterior probabilities of targets in multiple vision sensors is proposed.For the three visual senses,firstly,each visual sensor uses the classical particle filter method to calculate the weights of the particles in each visual sensor,and then calculates the epipolar lines of the particles in each visual sensor in other visual sensors.We get all the particles in the counterpoise that fall on the middle line with the opposite epipolar line as the midline,the height of the target box is the bandwidth,and the mean value of the weights of the particles falling in the polar band as the particles of the vision sensor in other vision The projection weights in the sensor,the particle weights and projection weights are multiplied to obtain the spatial weights of the particles,instead of the weights in the classical particle filter,the target is tracked.Aiming at the problem that the target similarity is easily affected by the illumination and other targets during the target tracking.this paper proposes a method to judge the target occlusion based on the deviation degree combined with the traditional similarity.First,we get the epipolar line of each particle in a camera in the other camera,then calculate the distance from each particle to the epipolar line in other cameras,then add the distance to all the distances as the target deviation of the camera.By judging the deviation and similarity of a target at a certain time,we can judge whether there is occlusion.Aiming at the problem of long time occlusion in target tracking,a multi vision cooperative target tracking method based on probabilistic model is proposed.When the target has been blocked by similarity and deviation,the opposite epipolar line in the camera's center of the camera is calculated and the particles are oscillated near the intersection point.Then the target state is estimated by the concussion particles,thus the multi vision sensor tracking is realized.Through a large number of experiments,the proposed multi vision cooperative target tracking method based on probability model can effectively solve the long time occlusion problem of the target,and has good robustness.