利用多特征混沌粒子滤波的视觉目标跟踪方法

随着摄像终端的增多以及自动视频分析需求量的增大，针对视频序列中存在突然运动、遮挡、运动模糊等干扰因素时传统视觉目标跟踪方法很难获得鲁棒性高、精确稳定的目标跟踪的问题，提出了利用多特征混沌粒子滤波的视觉目标跟踪方法。首先，基于非线性动力学预测进行混沌建模，利用混沌映射的梯度优化函数来搜索状态空间以找到参考轨迹；然后设计了一种用于视觉跟踪的混沌粒子滤波器，并改进运动表观模型，引入颜色、纹理和深度的特征完善滤波器的性能；最后，将多特征混沌粒子滤波器与其他视觉目标跟踪方法应用于VOT17和TB 数据集进行对比分析，以论证该方法的准确性。结果表明，提出的多特征混沌粒子滤波方法显著减少了粒子数量、搜索空间和滤波器发散，其精度高出其他方法约10%，在突然运动、遮挡和运动模糊等情况下整体性能优于其他几种对比方法。

Visual target tracking method using multi feature chaotic particle filter

With the increasing of camera terminals and demand for automatic video analysis, traditional visual target tracking methods have been difficult to obtain robustness, accuracy and stability to target tracking, when there are interference factors such as occlusion, illumination changes and motion blur in the video sequence. Aiming at the series of problems, this paper proposed a visual target tracking method based on multi-feature chaotic particle filtering. Firstly, this method carried out chaotic modeling based on nonlinear dynamics prediction, and used the gradient optimization function of chaotic map to search the state space to find the reference trajectory. Then this paper designed a chaotic particle filter for visual tracking and improved the sports appearance model, introduced features of color, texture and depth to improve the performance of the filter. Finally, it applied the multi-feature chaotic particle filter and other visual target tracking methods to the VOT17 dataset and TB dataset to demonstrate the accuracy of the proposed method. The results show that the multi-feature chaotic particle filtering method significantly reduces the number of particles, search space and filter divergence, and its accuracy is about 10% higher than other methods. It is superior to other methods in the case of sudden motion, occlusion and motion blur.