Desarrollo de un software de reconocimiento de errores en powerlifting usando visión artificial

Abstract

With the increase of users in gyms, the need for supervision of compound exercises such as powerlifts is increasing, since the execution of an incorrect technique can cause serious injuries to the athlete. In this project, a prototype web application was developed that uses computer vision to detect in real time errors such as incorrect hip placement, excessive back rounding and hyperextension at the end of the deadlift in powerlifting. The development methodology used followed a waterfall approach, structured in distinct phases: a document review to select intelligent algorithms; the development and integration of modules and the model; and the evaluation of the prototype in a controlled environment. Images of different variations of the power weight lifting technique were captured and labeled to train a model that could be run in web browsers. When evaluating the overall accuracy of the detector, it yielded a detection rate of 73.36%, a satisfactory performance in identifying critical errors. This system offers a versatile and robust solution for technique correction and power weight injury prevention; being a low cost and easy to implement alterna- tive, its use and application in real time will allow more businesses to use it. As future improvements, it is recommended to optimize the model to increase accuracy in the detection of less obvious variations.