A Human Motion Data Capture Study The University of Liverpool Rehabilitation Exercise Dataset

Abstract

The increasing accessibility of motion tracking technologies has resulted in a large amount of research focused on delivering exercise-based interventions remotely, coined under the term telerehabilitation. High quality human motion data is an essential component in the development and evaluation of Human Action Recognition (HAR) research, which plays a large role in exercise-based telerehabilitation. However, there is a lack of such human motion datasets for this domain, which hinders fast progress. This work presents a new human motion dataset named University of Liverpool Rehabilitation Exercise Dataset (UL-RED) containing 22 non-specialised exercises across 10 subjects and three data modalities: marker-based and marker-less motion tracking, and depth data. This dataset is the first to include motion repetitions of varying motion speeds, where subjects performed repetitions at a normal, fast, and slow pace. A total of 1,320 recordings were collected across the three data modalities, with over three hours of marker-based and marker-less motion tracking. This dataset is not only useful in the telerehabilitation landscape, but also within the wider field of HAR.