Research Of Object Tracking Based On Camshift Algorithm

Camshift tracking plays an important role in computer vision applications because of its robustness, ease of implementation and computational efficiency. It can perform well with objects that have a simple and constant appearance, while it is not robust in more complex cases. As it solely relies on back projected probabilities it can fail in cases when the object’s appearance changes, target and background have a large area of similar color or partial occlusion. In this paper, aiming at several cases existing in the object tracking algorithm, several schemes is proposed to resolve corresponding problems.(1)Aiming at several cases existing in the object tracking algorithm, such as will failure when the object’s appearance changes or when the target and background are similar, an object tracking algorithm based on multi-step color accumulated and texture information fusion is proposed. The whole tracking system can be divided into two parts:color and texture feature extracting system and target and background’s similarity judging system; in the tracking process those two parts run by turns according to different cases. When the object’s appearance changes, the Rol frame difference is used to compute the center of the target again and extracting multi-step color model and accumulate it to the old one.Tracking results show that the improved algorithm is robust to object’s appearance changes under complex background.(2)In order to solve the problem of occlusion,the author has deeply researched the particle filter. We address the issue of part-based tracking and employ an adaptive cue integration scheme. This is done by embedding the original tracker into a particle filter framework. Associating a reliability value to each fragment that describes a different part of the target object and dynamically adjusting these reliabilities at each frame with respect to the current frame context.The experimental results show that the algorithm pays the smaller time consumption for the higher tracking accuracy.(3)The author focus on the problem of similarly colored objects cross their trajectories. The tracking is likely to fail because more than one back-projected blob is present in the actual search region. To overcome this, we apply a hierarchical quad-tree re-detection strategy. In order to judge the target fail or not, we use double judgments as BH coefficient and histogram intersection.Tracking results show that this algorithm can lock the target when the target appears again after lost.We do a large number of simulation experiments, the experimental results and tracking error analysis show that the improved algorithm can obtain a higher level of accuracy than the existing algorithm and achieve real-time and efficient tracking.