Využití nositelných MOCAP ve fitness.

Abstract

This thesis explores the possibilities of using a wearable system for spatial position measurement to analyze and assess the correctness of performing fitness exercises. Selected exercises—the squat and Romanian deadlift—were used, and experimental measurements were designed with healthy volunteers who performed the chosen exercises under the supervision of a fitness trainer while being recorded by a wearable MOCAP (motion capture) system. Based on a literature review, the following parameters were statistically evaluated: knee, hip, and ankle flexion, knee adduction, and the duration of one repetition. To identify the correctness of the performed exercises, a model with a CNN-LSTM architecture in various configurations was used. Models using different numbers of sensors placed in various body locations were created. The model's accuracy was verified on a validation data set. Statistical differences were demonstrated between correctly and incorrectly performed exercises in the parameters of knee and hip flexion. The model used for classifying the correctness of exercise execution achieved comparable accuracy (test set accuracy above 99,8 %) using any single sensor placed on the lower body compared to other tested configurations. Validation on a separate data set reached an accuracy of 70 % for the squat and 30 % for the Romanian deadlift. The reduced accuracy on the validation data was caused by the simulation of correct and incorrect exercise executions.