Visual Tracking Using An Insect Vision Embedded Particle Filter

Particle filtering(PF) based object tracking algorithms have drawn great attention from lots of scholars. The core of PF is to predict the possible location of the target via the state transition model during the tracking process. One commonly adopted approach to propagate the sample set is resorting to prior motion cues under the smooth motion assumption, which performs well when the target moves with a relatively stable velocity. However, it would possibly fail if the target is undergoing abrupt motion(e.g. the motion of the football players in a match) or camera lens switching(e.g. the location of the target changes abruptly). To address this problem, inspired by insect vision, we propose a simple yet effective visual tracking framework based on PF. This algorithm can effectively solve the tracking problems when the target undergoes occlusion, abrupt motion and illumination changes.First, we propose a novel motion detect method to estimate the motion state of the target so as to refine the position state of propagated particles using more accurate transition mode. To address the uncertain motion problem, we introduce the neuronal computational model of the way biological ommateum processing information, where we devide the video frame into grids. Each grid is equal to one ommatidium and we can control the sensitivity of the detector via adjusting the size of the grid. Utilizing the output of the detector, we improved the transition model of the PF to refine a more accurate postion of the target.Furthermore, we design a novel sample optimization framework where local and global search strategies are jointly used to seek for the best candidate sample. Since local search method tries to find the best sample in a generative subspace and the global one in a discriminative subspace, they can benefit each other. Local search strategy is always used in each frame while golobal search strategy is triggered by two situations: one is for evry const frames; the other is when target lost occurs. As disappearance caused by long duration severe occlusion may lead to failure during the tracking, we propose a new method to monitor this situation. Once the target is lost, we could recover it via the global search method previously mentioned.Finally, we give an incremental learning based online template updating method to adapt the appearance changes.These ingredients lead to a new tracking algorithm. Experiments on some publicly available benchmark video sequences demonstrate that the proposed tracking algorithm outperforms the state-of-the art methods in challenging scenarios, especially for tracking target which is undergoing abrupt motion, target occlusion and scene illumination changes.