一种有效的体育视频目标跟踪算法

论文研究了面向体育视频的运动目标跟踪技术,提出了一种最优化的混合跟踪方法。首先,采用粒子滤波算法来预测运动目标的初略位置,通过比较预测位置上的目标同目标模型之间的相似度,当相似度小于一定的阀值时,认为目标运动模型发生了根本变化,需要启用新的运动模型;当相似度大于一定的阀值时则认为目标运动模型没有发生大的变化,不需要启用新的运动模型,通过这种方式找到目标的最优化运动模型。最后将最优化的运动模型用于基于核的均值转移算法中,从而获得运动目标的精确位置。

An Efficient Object Tracking Algorithm of Sports Video

We investigate the problem of object tracking from sports videos,and propose a new hybrid object tracking method.First,we use different dynamic model in particle filter algorithm to estimate the position of object,and then compute comparability between object model and object on the estimated position,if the comparability is less than const value,we consider the object's dynamic model has changed and need to modify the dynamic model;if the comparability is more than const value,we consider the object's dynamic model has not changed and don't need to modify the dynamic model,by this procedure,we can find the dynamic model of optimization.Finally we use this dynamic model of optimization in mean-shift algorithm,and compute object's true position.