基于单目视觉的FMS人体姿态识别

Human posture recognition based on monocular vision

针对部队日常体能训练中引入功能性动作筛查(FMS)的困难,提出一种基于单目视觉的自动化解决方案。通过改进现有人体姿态识别算法,再结合余弦相似性完成FMS动作提取。分析现有算法的弊端,提出利用帧间光流提高人体姿态识别精度的解决思路,并验证其可行性。

Aiming at the difficulty of introducing functional movement screening(FMS)in the daily physical training of troops,an automated solution based on monocular vision is proposed.The FMS action extraction is completed by improving the existing human body posture recognition algorithm and combining cosine similarity.The shortcomings of existing algorithms are analyzed and solutions to improve the accuracy of human posture recognition by using optical flow between frames are proposed,and the feasibility is verified.