LAUSR

Assessment of gait consistency requires testing over a long walking distance. Robot-mounted 3D cameras represent a cost-effective, markerless technology of human gait analysis that can be applied for this purpose. However, the use of robotic platforms for gait analysis is limited by the low accuracy of 3D cameras. The aim of this study is to improve the accuracy of kinematic and spatio-temporal estimations obtained from a robot-mounted 3D camera by applying a supervised learning process, and then to verify the effectiveness of the proposed method for clinical use. Artificial neural networks have been trained using the reference data provided by a Vicon system, which lead to improved estimations. Then, gait characteristics in Multiple Sclerosis patients has been measured. Significant differences respect to Healthy Controls have been found mainly for hip flexion, pelvis tilt, pelvis rotation and stride length. The improvement of robot-mounted 3D camera estimations and their application to the analysis of gait impairment in a natural environment show the flexibility and adaptability that this setup can provide.