Work

Genomförbarhet av träningsigenkänning på Exxentric kBox med IMU-sensor

Year: 2024
Type: dissertation
Abstract: Advancements in wearable technology have revolutionized the way physical activities are monitored. It offers new opportunities for exercise tracking and performance evaluation. The integration of Inertial Measurement Unit (IMU) data has emerged as a promising approach for exercise recognition, particularly in resistance training. Flywheel training, known for its ability to provide variable resistance, is increasingly utilized for both strength development and rehabilitation. However, despite the growing use of IMUs in physical activity monitoring, accurately recognizing the complex movements involved in flywheel training remains a challenge.To address this challenge, data were collected from 18 participants performing five of the most popular exercises in flywheel training—Squat, Deadlift, Bicep Curls, Bent-over Row, and High Pull. A comprehensive approach was employed which involves data cleaning, synchronization, and advanced machine learning techniques, including both classic and deep learning models, toaccurately recognize these exercises. Sensor placement and combinations were also analysed to optimize model performance and explore wearable solutions for real-world implementation.The results demonstrated that the CNN model achieved 94.36% accuracy when using data from all four sensors placed on the wrist, thigh, arm, and ankle, making it the best-performing model. Further analysis revealed that optimizing the number of sensors could yield even better results. Specifically, the combination of wrist, thigh, and arm sensors provided the highest accuracy of 98.08%. This indicates that three strategically placed sensors can outperform the initial four-sensor setup. Additionally, the wrist sensor alone proved to be the most effective single sensor, with an accuracy of 87.36%. These findings underscore the importance of strategic sensor placement and highlight the potential of deep learning techniques to enhance exercise recognition in flywheel training.However, the study encountered several challenges. One challenge was the need to create a selfcollected dataset due to the lack of publicly available data. Additionally, there were difficulties in distinguishing between exercises with similar movement patterns using a single sensor, particularly in exercises like the deadlift and bent-over row. These results highlight the need for further research into advanced data processing techniques and the expansion of datasets to improve recognition accuracy.This research provides valuable insights into the application of IMUs in flywheel exercise recognition, with implications for personalized exercise monitoring and the potential for betterrehabilitation and training experiences.
Author: Yuping, Huang
Language: English
Open Access status: green
OpenAlex ID: W7009973425
Cites: 36
Cited by:
Related to: 10