运动动作图像正确姿态轮廓提取方法对比研究
2018-08-06邹文君肖永君
邹文君 肖永君



摘 要: 由于目标轮廓提取方法一直存在正确姿态轮廓提取不准确的问题,提出两种新的运动动作图像正确姿态轮廓提取对比方法。通过相邻帧差法对运动动作图像正确姿态区域进行提取,在此基础上介绍了snake模型方法,其将运动动作图像正确姿态轮廓的提取问题转换成求解能量函数的极小化问题。分析了水平集方法,将演化曲线隐性地描述成运动动作图像上的三维连续函数曲面的零水平集,当水平集函数为0时,得到的结果即为轮廓提取结果。通过实验对比所提两种方法的有效性,发现snake模型方法的轮廓提取效果、完成性和信息丢失程度虽然优于其他方法,但劣于水平集方法。
关键词: 运动动作提取; 动作图像; 能量函数; 水平集; 正确姿态; 轮廓提取
中图分类号: TN911.73?34; TP391.41 文献标识码: A 文章编号: 1004?373X(2018)15?0065?04
Comparative study on extraction methods of correct pose contour in moving action image
ZOU Wenjun, XIAO Yongjun
(Institute of Technology, East China Jiaotong University, Nanchang 330100, China)
Abstract: The traditional method of target contour extraction has the problem of incorrect pose contour extraction. Two new comparative methods of correct posture contour extraction for moving action image are proposed, in which the adjacent frame difference method is used to extract the correct pose region of the moving action image. The snake model method is introduced to convert the correct pose contour extraction from moving action image into the solving of energy function minimization. The level set method is analyzed, in which the evolution curve is implicitly described as the zero level set of the 3D continuous function surface on the moving action image. When the level set function is zero, the obtained result is deemed as the contour extraction result. By comparing the effectiveness of the two methods, it is found that the snake model method is better than other methods in the aspects of contour extraction effect, completeness and information loss rate, but is inferior to the level set method.
Keywords: moving action extraction; motion image; energy function; level set; correct posture; contour extraction
0 引 言
目標轮廓提取就是在一段视频序列的每帧图像中精确找到所感兴趣的运动目标的轮廓[1?2]。随着运动比赛竞争的越发激烈,运动员对自身动作准确度的要求越来越高[3?4]。因此,研究一种轮廓提取方法对运动动作图像正确姿态轮廓进行提取具有重要意义,能够为运动员的训练提供技术支持[5?6]。
文献[7]提出一种离散可变模型对目标轮廓进行提取,通过图像的外部约束令轮廓曲线保持在图像边缘处,通过图像内部约束对轮廓曲线形状进行调整。该方法实现过程简单,但提取的轮廓结果不可靠。文献[8]提出一种基于带电……
