武术难点动作图像多目标自动提取方法研究

Research on Multi-target Automatic Extraction Method of Difficult Action Images in Wushu

提取武术难点动作时常常因无法消除动作图像中的阴影部分,导致存在角点检测重复率高、提取效果差和整体识别率低的问题。为此,提出武术难点动作图像多目标自动提取方法,建立武术动作图像的背景模型,并更新背景模型,通过背景差分与分割处理获取武术动作图像的目标区域。检测目标区域中是否存在阴影,并且消除阴影,检测武术动作角度,根据角度数据完成武术难点动作提取。实验结果表明,所提方法的角点检测重复率低、提取效果好、整体识别率较高。

When extracting difficult martial arts movements,the shadow part of the action image can not be eliminated,which leads to the problems of high repetition rate of corner detection,poor extraction effect and low overall recognition rate.In order to solve the above problems,the paper proposes a multi-objective automatic extraction method for martial arts difficult action images,establishes the background model of martial arts action images,updates the background model,and obtains the target area of martial arts action images by background difference and segmentation.The shadow is detected in the target area,and the shadow is eliminated,the angle of martial arts action is detected,and the difficult action extraction of martial arts is completed according to the angle data.The experimental results show that the proposed method has low repetition rate,good extraction effect and high overall recognition rate.