Rope Jumping Strength Monitoring on Smart Devices via Passive Acoustic Sensing

Abstract

Rope jumping, as a fitness exercise recommended by many sports medicine practitioners, can improve cardiorespiratory capacity and physical coordination. Existing rope jump monitoring systems have limitations in terms of convenience, comfort, and exercise intensity evaluation. This paper presents a rope jump monitoring system using passive acoustic sensing. Our system exploits the off-the-shelf smartphone and headphones to capture the user’s rope-jumping sound and breathing sound after exercise. Given the captured acoustic data, the system uses a short-time energy-based approach and the high correlation between rope jumping cycles to detect the rope-jumping sound frames, then applies a dual-threshold endpoint detection algorithm to calculate the number of rope jumps. Finally, our system performs regression predictions of exercise intensity based on features extracted from the jumping speed and the mel spectrograms of the user’s breathing sound. The significant advantage of the system lies in the solution of the problem of poorly characterized mel spectrograms. We employ an attentive mechanism-based GAN to generate optimized breathing sound mel spectrograms and apply domain adversarial adaptive in the network to improve the migration capability of the system. Through extensive experiments, our system achieves (on average) 0.32 and 2.3% error rates for the rope jumping count and exercise intensity evaluation, respectively.

Keywords

• acoustic sensing
• domain adversarial adaptation
• generative adversarial network
• passive rope jump monitoring
• short-term energy