基于自适应粒子滤波的跳水运动视频跟踪算法

A Video Tracking Algorithm for Diving Based on Adaptive Particle Filtering

用传统粒子滤波算法对跳水运动视频跟踪存在两个突出问题:观测模型不能适应运动员身体的表观变化;运动模型不能准确预测运动员位置的快速改变。针对这两个问题,本文提出一种自适应粒子滤波算法。该算法在粒子滤波框架下引入一种自适应观测模型,并且根据跟踪误差与运动员动作改变幅度的大小,自适应选择噪声方差和粒子数量。实验结果表明,本文算法比传统粒子滤波算法具有更低的跟踪误差率,而且在运动员动作改变幅度变大时有更好的鲁棒性。

Video tracking of diving with the original particle filtering aglorithms remain two major problems. First the observation model can not adapt to the changes of athlete body appearance. Second the fixed noise variance in the kinematics model usually leads to a failure to predict the rapid changes of the athlete location. We propose an adaptive particle filtering algorithm to tackle these problems. This algorithm introduces an adaptive observation model under the particle filtering framework, and adaptively chooses the noise variance and the particle quantity according to the errors of tracking and the athletes＇ act change margin. The experimental results show that the algorithm proposed in this paper has a lower tracking error rate compared to the traditional particle filtering algorithms, and has better robustness when the athletes＇ act change margin is bigger.