The paper discusses the challenges and advancements in location-invariant and device-agnostic motion activity recognition on wearable devices. Researchers highlight the problem of sensing heterogeneity, which requires custom models for different platforms. They present comprehensive evaluation of motion models across sensor locations and introduce the largest multi-location activity dataset. The team also presents deployable on-device motion models with high accuracy, irrespective of sensor placements. The study investigates cross-location data synthesis to lessen the burden of data collection tasks. The findings advance the vision of low-barrier, location-invariant activity recognition systems, promoting research in HCI and ubiquitous computing.
Publication date: 2024-02-06
Project Page: https://arxiv.org/abs/2402.03714
Paper: https://arxiv.org/pdf/2402.03714
