Benchmarking Deep Learning-Based Pose Estimation from Sparse IMUs
Human pose estimation from a small number of body-worn inertial measurement units (IMUs), such as accelerometers and gyroscopes, holds great potential for practical applications in areas like augmented and virtual reality (AR/VR) and healthcare. While recent advances in deep learning have significantly improved the accuracy of such methods, the lack of standardization in training protocols, evaluation metrics, and datasets makes consistent benchmarking and comparison difficult. This project aims to address this gap by developing a standardized benchmark and evaluation platform for sparse IMU-based pose estimation methods. The platform will include automated evaluation tools and a simple web interface. Ideally, it will be accompanied by a newly recorded motion benchmark dataset. In the case of exceptional progress, the project may lead to a publication.