A 3D camera system for gait speed assessment

The gait speed test is one of the most widely used ways to assess frailty in older adults. Traditionally, the test is performed by walking at normal speed while a healthcare practitioner records the time it takes to cover a previously set distance using a stopwatch. This work proposes a system for frailty assessment that initially focuses on the gait speed test but is designed to be extendable to other functional tests, such as the Short Physical Performance Battery (SPPB). The system combines efficiency for healthcare professionals with a user-friendly experience for patients. To validate the proposed system, our study compares the typical human error in conventional gait speed testing with the accuracy of a 3D camera-based system using the photo-diodes as ground truth. We conducted a total of 571 walks in both real and laboratory environments and a total of 26 subjects, including older adults, participated in the study. In the four experiments carried out, the single-camera system achieved MAE (time) of 43.2–129.4 ms, RMSE 68.3–199.3 ms, MAPE 1.62–3.86%, and PCC 0.89–0.99 versus photodiodes; in speed terms, MAE was 1.53–3.18 cm/s. The worst-case speed MAE (3.18cm/s) improves on a single-camera competitor (Yagi et al. 3.62cm/s) by 0.44 cm/s ( ≈ 12 % ) , and on a three-camera Azure Kinect system (Arizpe-Gómez et al. 4.6 cm/s) by 1.42 cm/s ( ≈ 31 % ) . • A single 3D camera based system for gait speed measurement is proposed. • In all the experiments, the system achieves a significantly smaller MAE than humans. • The highest (and thus the worst) MAE in speed obtained is 3.18 cm/s. • A total of 571 walks were carried out in both real and laboratory environments.