基于卡尔曼预测粒子滤波的网球运动目标跟踪方法

Tennis Motion Tracking Based on Particle Filter in Kalman Filter Prediction

为有效跟踪视频网球运动,提出了一种基于卡尔曼滤波预测的粒子滤波网球运动跟踪方法。基于多尺度小波变换在时域和空域均具有优异的局部化特征,将相邻帧视频图像进行差分,提取反映前景运动的目标特征信息,克服光照变化以及网球运动尺度随时不断变化的不利因素影响;同时,基于网球场地结构化特性,排除场地外不利干扰因素影响。在此基础上,采用卡尔曼滤波对粒子进行预测和修正,将当前观测信息融入到粒子滤波过程中,估计预测粒子状态的均值和协方差,使动态粒子更加接近其后验概率分布,从而提高网球运动目标的跟踪精度。通过与同类方法在不同网球公开赛的定量对比,实验结果表明,所提方法能有效跟踪视频网球运动目标。

A tennis motion tracking method has been proposed based on particle filter in Kalman filter prediction from the video sequence.Some motion foreground information has been extracted from the difference between consecutive frames according to outstanding localization characteristics both in temporal and spatial domains for multi-scale wave transformation.Some adverse factors are overcome including illumination variation,dynamic changing in scale of tennis motion over time.Besides some structural characteristics of the tennis courts are employed to exclude efficiently some disturbing influences from outside the site.Kalman filter is used to predict and correct some particles states during particle filtering.Some current measurement information is combined into the particle filter processing.Both mean value and covariance for predicting the particle state are estimated to make the distribution of dynamic particle get close to its posterior probability.The tracking result for the tennis motion is improved.The tracking performance is compared with some similar methods quantificationally based on some different tennis opens.Experimental results have been shown that the performance of the developed method outperforms the investigated methods.It can be applied to track the tennis motion from video efficiently.