Research On Target Tracking In Panoramic Vision

Compared with the conventional common perspective imaging camera, panoramic vision has 360 degrees of field of view, so it is widely used in robot navigation, virtual reality, video surveillance and other fields have great prospects for development. Because of the resolution of the panorama image distribution is very uneven, and the movement of the target in the panoramic image can present a violent rotating phenomenon, The classic particle filter algorithm has poor effect for the panoramic image of the target tracking. In addition, A single camera can only get the target information in a direction, the target information is not comprehensive. So it cannot handle the target occlusion problem. In this paper, the target tracking method for panoramic vision is proposed.In the improved particle filter algorithm for panoramic vision, At first, the single viewpoint catadioptric panoramic imaging system is introduced. Based on the principle of panoramic imaging, the panoramic image coordinate system is mapped into a unit sphere coordinate system. Then defined the target state on the unit sphere, and then use the particle filter on the spherical This makes the algorithm can solve the problem of target tracking failure due to distortion and rotation of the panoramic image target. In the two panoramic vision sensor collaborative target tracking, firstly this paper introduces the epipolar geometry theory for the common vision. and then deduces the epipolar geometry theory for the panoramic vision.Secondly,do research on the collaborative target tracking mechanism with two panoramic vision. Considering that the resolution distribution of the panoramic image is very uneven, a novel camera collaboration likelihood model for panoramic vision is proposed. Finally, achieve the collaborative target tracking in panoramic vision.Through a large number of experiments show that the proposed algorithm of the improved particle filter algorithm for panoramic vision can effectively solve the problem of distortion and rotation. The proposed algorithm of the two panoramic vision sensor collaborative target tracking can effectively solve the problem of target occlusion.