提出一种适用于跳水运动视频的全局运动估计方法。跳水运动中的运动员着装简单、肤色遮挡较少,通过测试闽值建立肤色模型,快速提取前景和背景中的人体区域,避免其对全局运动参数估算的影响;利用帧差法去除相邻图像帧内相对静止的光流场,根据参照宏块运动方向进一步去除影响全局运动估计的宏块,减少宏块匹配时间。实验结果表明,算法不但简化了全局运动参数估算的复杂性,而且提高了的计算精度。
This paper presents a global motion estimation method for diving video. Human body color model is constructed base on the athletes' simple dress in diving video and the area of human body within the foreground and background is picked up so as to improve the efficiency of global motion estimation. Meanwhile, the adjacent frame difference and the direction of the reference module are used to remove the modules which not used in the global motion estimation, so module-matching time is reduced accordingly. Experimental results show that the presented method can not merely reduce the computational complexity of the global motion estimation's, but also improve its accuracy.