摘要
针对体育视频中运动目标的关键姿态不一,人工提取视频图像中的关键姿态费时费力的问题,提出一种基于深度学习的关键姿态提取算法。利用距离变换根据体育视频中像素点特征,对关键姿态进行判断,分别运用骨架候选点和背景候选点,计算体育视频中运动目标的关键姿态。实验表明,提出的算法能够准确清晰提取视频中的运动目标特征轮廓,并能够区分关键姿态点和轮廓点,从而实现体育视频的人体关键姿态的提取。
Based on the problem that the key posture of moving targets are different in sports video and it is time-consuming and laborious to manually extract the key postures in video images,a key pose extraction algorithm based on deep learning is proposed.According to the characteristics of pixels in sports video,the key posture is judged by distance transformation,and the key posture of moving targets in sports video is calculated by using skeleton candidate points and background candidate points respectively.Experiments show that the proposed algorithm can accurately and clearly extract the moving target feature contour in the video,and can distinguish the key posture points and contour points,so as to realize the extraction of human body key posture in sports video.
作者
张婷
于涛
王欣
ZHANG Ting;YU Tao;WANG Xin(Department of Physical Education,Liaoning University of Traditional Chinese Medicine,Shenyang 110085,China)
出处
《信息技术》
2023年第5期1-5,12,共6页
Information Technology
基金
国家自然科学基金(61502125)
辽宁中医药大学基金(2019-lnzy028)。
关键词
体育视频提取
视频序列
深度学习
关键姿态
距离变换
sports video extraction
video sequence
deep learning
key posture
distance transformation
