A machine vision-based method for recognizing standard movements in college sports tests

Abstract

Currently, the physical health of college students is extremely concerning, as a significant number of them lack physical activity, resulting in a severe decline in physical health. To promote the physical health of students and ensure the health of all people, the state has issued a policy to include the physical health of college students in the assessment and evaluation of colleges and universities, and the physical health test of college students has become an important criterion in the assessment and evaluation of different colleges and universities. However, because many test movements must be evaluated and scored by professional coaches, the formalization and marginalization of sports tests in colleges and universities has resulted from the lack of professional sports test coaches and teachers. To increase the professionalism of sports test instructors and coaches, as well as to reduce the labor costs associated with sports tests. In this paper, we propose using machine vision to recognize the sports test actions of college students by extracting the depth information of the actions via vision sensors, combining the information of human skeletal nodes, and separating the action image information from the background by choosing an appropriate depth distance threshold. Using a support vector machine (SVM) algorithm, the segmented gesture images are recognized and their normality is assessed. The conclusion of the dynamic action recognition and evaluation process is reached. Extensive testing demonstrates that this method is able to recognize test actions in real time and provide accurate scores.