非刚性人体运动动作图像姿态轮廓提取算法

针对人体运动动作图像提取传统方法仅能获取局部最佳解,提取的特征序列不连续,导致轮廓提取效果差的问题,提出一种非刚性人体运动动作图像姿态轮廓提取算法.首先,对人体运动动作图像序列中的尺度不变特征变换(SIFT)进行提取预处理,获取人体特征提取图,设计一种人体运动序列顺序概率图模型,以保证特征序列提取的连续性;其次,构建人体肢体外观模型,基于该外观模型采用序列影像高精度轮廓提取算法提取当前人体运动动作帧的轮廓线.实验结果表明,该算法能提取连续的特征序列,提取的人体姿态轮廓精确度较高,且具有较高的效率和鲁棒性.

Image Attitude Contour Extraction Algorithm of Nonrigid Human Motion

Aiming at the problem that the traditional methodof image extraction of human motion could only obtain the local optimal solution, and the extracted feature sequence was discontinuous, which led to the poor contour extraction effect, we proposed an image attitude contour extraction algorithm of nonrigid human motion. First, the scale invariant feature transform (SIFT) in the human motion image sequence was extracted and preprocessed to obtain the human feature extraction graph, and a sequential probability graph model of the human motion sequence was designed to ensure the continuity of the feature sequence extraction. Then the exterior appearance model of human limbs was constructed. Based on the appearance model, the high precision contour extraction algorithm of sequential image was used to extract the contour line of the current human motion frame. The experimental results show that the proposed algorithm can extract continuous feature sequences, and the extracted human pose contours have high accuracy, high efficiency and robustness.