Abstract
Gait parameters, including the step length, often reflect the functional states of patients. Several years ago, Microsoft released a markerless motion capture system named Kinect, which attracted attention because of its potential for the assessment of human motion in clinical settings. However, the skeleton information produced by Kinect has been found not accurate enough for the rehabilitation purposes. The skeleton of Kinect is produced from depth images captured with the infrared camera. Therefore, if the depth images are analyzed with custom-made software, the accuracy of gait parameters may be improved. To explore this possibility, the accuracy of a spatial gait parameter estimated from the depth images was evaluated by comparing it with that obtained by the Kinect skeleton method. Ten young healthy male subjects walked on a treadmill. The step length was calculated as a spatial gait parameter simultaneously from the depth images and the Kinect skeleton. The estimates were compared to the reference values measured with a marker-based three-dimensional motion capture system. As a result, the root mean square error of the step length, which represented the variability against the reference values, was significantly smaller for the depth images method than for the skeleton method. Thus, by analyzing the depth images, it may be possible to build a motion capture system that has higher accuracy than the Kinect skeleton method, and, at the same time, is more convenient than the conventional marker-based motion capture system.
