Fine-Fit: A Fine-grained Gym Exercises Recognition System
Zhe Zhang, Nian Wang, Li Cui · 2018
Gym exercise recognition can help people to monitor and supervise their personal exercise progress. Although previous works have already recognized gym exercises by using wearable or external sensors, they failed to cover the recognition of activities at a finer level which means to identify exercises that perform similar activities but are significantly different in muscle firing. These finer level exercises, defined as fine-grained gym exercise in this paper, can cause completely different training effects to exercisers including muscle strength building, muscle circumference shaping, etc. The goal of this work is to propose a novel fined-grained gym exercise recognition system, namely Fine-Fit, which is aiming to provide more detailed sports information useful in monitoring exercises, assessing non-standard actions, and avoiding muscle injury as well. The unique design of Fine-Fit is to use a single source sensor (accelerometer) to sense the body movement and the corresponding muscle vibration simultaneously (Single Source Multi-view Sensing method, SSMS) for high accurate fine-grained exercise recognition. Besides, Fine-Fit proposes a novel feature, namely Muscle Recruitment Energy Coefficient (MREC), particularly for fine-grained exercise classification. MREC implicitly reflects the correlation of elastic potential energy and the corresponding moving distance of muscles and it is able to improve the accuracy of identification effectively. The preliminary results demonstrate that Fine-Fit can recognize fine-grained gym exercises of push-ups and barbell curls with different bearing weights and hands distances at 91% precision and Fine-Fit enhances activity recognition to a fine-grained level.