基于颜色的快速人体跟踪及遮挡处理

为了对被跟踪到的运动员进行运动姿态以及运动参数的分析，给运动员的训练提供科学合理的参考，提高比赛成绩，研究了面向体育视频的运动目标跟踪技术，提出了一种基于Mean Shift的综合算法.首先，根据背景加权直方图选择跟踪目标与背景图像的差别最显著的部分作为跟踪特征，以减少背景信息对跟踪效果的影响；其次，针对Mean Shift算法需要对图像进行穷举匹配的问题，利用Kalman滤波对目标的状态进行有效预测，减少了匹配运算次数，改善了快速运动目标的跟踪效果，提高了跟踪算法的稳健性；最后运用基于核的Mean Shift算法对运动目标进行跟踪，同时进行目标模板的实时更新，实现了对体育视频中运动员的稳定实时的跟踪.该方法成功地解决了部分遮挡、背景混乱以及目标尺寸变化等问题.

Tracking fast movement using colors while accommodating occlusion

A new integrated algorithm was proposed to improve tracking of fast movement in sports videos. First, from the weighted background histogram, the portion of the image with the biggest differences between the tracking target and the background was selected as the tracking feature. This reduced the influence of background information on tracking effects. Next, since the mean shift algorithm needs exhaustive matching with images, we used Kalman filtering to effectively predict the target’s state. This decreased the number of matching calculations and thus improved the robustness of the tracking algorithm. A kernel based Mean Shift algorithm was then employed to track moving targets. Finally, real time updating of the target template was provided. The method ensures stable, real time tracking of athletes in sports videos. Also, other problems, such as partial occlusions, chaotic backgrounds and variation in target size were successfully dealt with using the proposed algorithm.