iMove: Exploring Bio-impedance Sensing for Fitness Activity Recognition

TL;DR

iMove tackles fitness activity recognition by integrating bio-impedance sensing with IMU-based HAR through sensor fusion and a cross-modality contrastive learning framework. The method enables BI information to train better IMU representations, allowing IMU-only deployment, while BI+IMU fusion yields the strongest performance (upper-body Macro F1 of $89.57\%$, lower-body $81.74\%$). In upper-body experiments, BI alone achieves $75.36\%$ and IMU alone $81.49\%$, with IMU-trained BI transfer achieving $84.71\%$ Macro F1; the approach also generalizes to lower-body activities in an extended study. These findings indicate the practical potential of BI as a training-time auxiliary and of sensor fusion for enhancing HAR in wearable systems, paving the way for robust, privacy-preserving fitness tracking.

Abstract

Automatic and precise fitness activity recognition can be beneficial in aspects from promoting a healthy lifestyle to personalized preventative healthcare. While IMUs are currently the prominent fitness tracking modality, through iMove, we show bio-impedence can help improve IMU-based fitness tracking through sensor fusion and contrastive learning.To evaluate our methods, we conducted an experiment including six upper body fitness activities performed by ten subjects over five days to collect synchronized data from bio-impedance across two wrists and IMU on the left wrist.The contrastive learning framework uses the two modalities to train a better IMU-only classification model, where bio-impedance is only required at the training phase, by which the average Macro F1 score with the input of a single IMU was improved by 3.22 \% reaching 84.71 \% compared to the 81.49 \% of the IMU baseline model. We have also shown how bio-impedance can improve human activity recognition (HAR) directly through sensor fusion, reaching an average Macro F1 score of 89.57 \% (two modalities required for both training and inference) even if Bio-impedance alone has an average macro F1 score of 75.36 \%, which is outperformed by IMU alone. In addition, similar results were obtained in an extended study on lower body fitness activity classification, demonstrating the generalisability of our approach.Our findings underscore the potential of sensor fusion and contrastive learning as valuable tools for advancing fitness activity recognition, with bio-impedance playing a pivotal role in augmenting the capabilities of IMU-based systems.

Figures (16)

• Figure 1: Hardware design and prototype of iMove
• Figure 2: Upper body classification results with different experiment configurations (stimuli frequency and electrode location)
• Figure 3: Bio-impedance and IMU signal of six upper body fitness activities (The IMU signal is the L2-Norm of raw data. Chest Press: participant holds a one-kilogram dumbbell in each hand and presses the dumbbells away from the chest until arms are fully extended, and brings them back to the starting position; Shoulder+Chest Press: participant adds an action of lifting the dumbbells from shoulder height to above his/her head when his/her arms are fully extended based on chest press 1 activity; Arm hold+shoulder press: participant holds a dumbbell with an overhand grip at shoulder height by one hand, push the dumbbell overhead by fully extending his/her arm, while keeping the other arm at shoulder height, change the hand to hold the dumbbell after more than ten times)
• Figure 4: t-SNE plots of six upper body fitness activities with different sensing modalities
• Figure 5: Backbone architecture of the neural network used for classification (Backbone)