The Contribution of Human Body Capacitance/Body-Area Electric Field To Individual and Collaborative Activity Recognition

Sizhen Bian,Vitor Fortes Rey,Siyu Yuan,Paul Lukowicz

DOI: https://doi.org/10.48550/arXiv.2210.14794

IF: 4.729

2022-10-26

Signal Processing

Abstract:The current dominated wearable body motion sensor is IMU. This work presented an alternative wearable motion-sensing approach: human body capacitance (HBC, also commonly defined as body-area electric field). While being less robust in tracking the posture and trajectory, HBC has two properties that make it an attractive. First, the deployment of the sensing node on the being tracked body part is not a requirement for HBC sensing approach. Second, HBC is sensitive to the body's interaction with its surroundings, including both touching and being in the immediate proximity of people and objects. We first described the sensing principle for HBC, sensor architecture and implementation, and methods for evaluation. We then presented two case studies demonstrating the usefulness of HBC as a complement/alternative to IMUs. First, we explored the exercise recognition and repetition counting of seven machine-free leg-only exercises and eleven general gym workouts with the signal source of HBC and IMU. The HBC sensing shows significant advantages over the IMU signals in classification(0.89 vs 0.78 in F-score) and counting(0.982 vs 0.938 in accuracy) of the leg-only exercises. For the general gym workouts, HBC only shows recognition improvement for certain workouts like adductor where legs alone complete the movement. And it also supplies better results over the IMU for workouts counting(0.800 vs. 0.756 when wearing the sensors on the wrist). In the second case, we tried to recognize actions related to manipulating objects and physical collaboration between users by using a wrist-worn HBC sensing unit. We detected collaboration between the users with 0.69 F-score when receiving data from a single user and 0.78 when receiving data from both users. The capacitive sensor can improve the recognition of collaborative activities with an F-score over a single wrist accelerometer approach by 16\%.