Human Pose Estimation based Speed Detection System for Running on Treadmill
Zhonghan Zhao, Shanzhen Lan, Shujun Zhang · 2020
Nowadays, with the improvement of life quality, people gradually realize the importance of mass entertainment and physical fitness. A growing number of people watch sports events and plan to run together in their leisure time, but the existing researches on human running speed detection always ignored the characteristics of the human body. For complex detection, auxiliary equipment needs to be worn during detection, and intelligent detection cannot be achieved. Therefore, we have designed a system that intelligently detects the running speed of people, provides a platform for runners to monitor and share their running status in real-time, and is an auxiliary means for sports events. In this paper, we design a human pose based system that extracts key point information from images to detect the speed with high precision. Additionally, based on this system, we realize intelligent detection while solving the detection of the relatively static position of the human body. Data were collected while the participant was walking/running at different speeds on a treadmill. In the experiment, the key point accuracy of the human pose estimation system based on the Simple Baselines combined with the MPII data set is 89.2%. The model can meet the accuracy requirements of speed detection. In the actual operation process, the system completed the detection task. In the experiment, the average relative error of speed sequences is 4.89%.