面向引体向上的视频图像关键点检测监测方法

Key-point detection of video picture monitoring method for pull-up

引体向上是军队有效的力量训练科目,对引体向上进行监测能提高训练效率。为实现引体向上的自动监测,分析了引体向上动作,确定了以下颌与手臂为主的关键点检测目标,提出一种用手部高度代替单杠轮廓进行检测的新方法。改进了Openpose网络,设计了关键点检测网络模型,并使用自构建数据集进行训练;使用YOLOv5与自构建网络结合构建双人监测网络模型,识别双人引体向上;与其他方法进行对比,该方法具有运算量小和能够双人识别的优点,可保持94%的准确率,达到了较好的监测效果。

Pull-up is an efficient subject of military strength training.Monitoring pull-up exercise can improve fighting capacity.In order to achieve automatic monitoring,we firstly analyzed the action of pull-up training,and chose the chin and arms as our target of key-point detection.We proposed a new method to detect pull-up training instead of identifying the horizontal bar.Secondly,we designed a key-point detection network based on Openpose,and trained it with self-built dataset.Thirdly,we combined YOLOv5 with our network to build a double-people-monitoring network model which can distinguish the double pull-ups.Lastly,we compared our experimental results with other methods,and draw a conclusion that our model has the advantages of small amount of calculation and double pull-up identification with 94%accuracy.