Gait analysis on daily data using IMUs in smart phones, watch and earbuds

The use of inertial measurement unit (IMU) sensors for gait analysis has become more prevalent. More options are being offered under the development of wearable smart devices and smartphones. These techniques provide a cost-effective way to collect motion data from everyday activities, addressing the limitations of controlled laboratory environments. Despite the potential of these technologies, there are still many challenges in analyzing gait data from everyday life.

Experiments involved 16 participants (7 women, 9 men; mean age: 27.69 years) who performed walking, jogging, and going up and down stairs under three smartphone-carrying conditions: pocket, backpack, and shoulder bag. Data were collected using iPhone 14, Apple Watch Series 10, and AirPods Pro, supplemented with Xsens motion capture for ground truth. IMU data from accelerometers and gyroscopes were preprocessed and standardized before applying Principal Component Analysis (PCA). A novel sliding window-based algorithm was developed for gait segmentation and grouping, featuring a Continuity-Matching Score (CMS) for evaluating both continuity and match quality.

The proposed algorithm achieved an overall segmentation accuracy of 89.25%, with the highest performance (90.38%) observed when the smartphone was carried in a pocket. Rand Index values confirmed reliable gait grouping, with minor accuracy reductions under more dynamic carrying conditions, such as backpacks. For walk-only dataset, segmentation accuracy improved to 95.67%, while for run-only dataset, the accuracy reached 96.21%.

This study introduced a system for daily-life gait analysis using consumer-grade IMU-equipped devices. The algorithm is capable of handling data containing multiple gait types, achieving reliable segmentation and grouping of synchronous gaits. Future work will focus on enhancing algorithm adaptability to dynamic environments and expanding its applicability to larger and more diverse datasets.