基于Mean Shift的目标跟踪算法性能比较研究

Performance evaluation of target tracking algorithms based on mean shift

针对目前尚未有对基于Mean Shift的各类目标跟踪算法在同一数据集下进行性能比较这一问题,选取了七种具有代表性的基于Mean Shift的目标跟踪算法,分别从算法时效性、跟踪成功率和跟踪精确度三个方面对算法在25段包括多种复杂场景的图像序列上的性能进行了仿真实验比较。实验结果给出了算法在不同评价指标以及不同图像场景下的性能表现。由实验得出的结论可以为基于Mean Shift的目标跟踪算法的进一步优化改进提供参考。

For the problem that there has not been the performance evaluation for kinds of target tracking algorithms based on mean shift under the same data sets,seven representative target tracking algorithms based on mean shift are selected to be evaluated in this paper. The tracking performance is evaluated experimentally on 25 video fragments involving kinds of complex scenes from time-efficiency,success rate and precision. Experiment result shows tracking performances are given under different evaluation indexes and different scenes. The conclusion will provide a reference for the optimization and improvement of target tracking algorithms based on mean shift in the future.