摘要
武术动作姿势包括静态与动态,呈现多维性,提出基于机器学习的武术动作姿势识别方法。采用小波阈值变换去除原始图像噪声干扰,参考运动过程中关节角度变化曲线,提取人体静态特征和动态特征,构建动作图像多维分割模型,结合机器学习与目标姿势参数,获取武术动作姿势边缘轮廓的特征分布函数,完成武术动作识别。实验证明:动作识别精准度较高、特征分辨能力较强,具有很好的姿势检测和辨识能力,能够满足武术训练对姿势细节获取要求。
Martial arts posture include static and dynamic,presenting multi-dimensionality,and a method of martial arts posture recognition based on machine learning is proposed.The wavelet threshold transformation is used to remove the noise interference of the original image,and the static and dynamic characteristics of the human body are extracted through referring to the change curve of the joint angle during the movement,to construct a multi-dimensional segmentation model of the posture image.Combining machine learning and target posture parameters,the feature distribution function of the edge contour of martial arts posture is obtained,and the recognition of martial arts posture is completed.Experiments have proved that posture recognition has high accuracy,strong feature resolution,and good posture detection and recognition capabilities,which can meet the requirements of martial arts training for obtaining posture details.
作者
高雅男
GAO Ya-nan(Shaanxi University of Science&Technology,Xi’an 710021,China)
出处
《信息技术》
2023年第2期30-34,40,共6页
Information Technology
基金
陕西省教育厅专项科研计划项目(19JK0122)。
关键词
小波阈值变化
关节角度变化曲线
动态特征
图像多维分割
机器学习
wavelet threshold change
change curve of the joint angle
dynamic characteristics
multi-dimensional image segmentation
machine learning
