Vision-Based Stair Recognition and Feature Extraction Method for Lower-Limb Wearable Robots

This paper proposes a real-time stair recognition and feature extraction method that can be used with lower limb wearable robots such as orthoses and prostheses. It primarily aids individuals with stereopsis disorder, which affects human depth perception, creating imbalances during day-to-day locomotion activities. Stair climbing is one of the most common human locomotion activities that require proper depth perception. This system uses a depth camera to obtain the step height and length information of the stair and generates a point cloud and a novel algorithm to process the data. An inertial measurement unit is externally attached to the camera to obtain the orientation data. The camera is mounted to the lateral side of the thigh after analyzing different mounting positions such as waist, thigh, and shank to identify the optimal configuration. The experimental results for step length and height were obtained and compared with the actual step sizes. The average accuracy of the proposed algorithm is higher than 98 % for indoor environments and 97 % for outdoor environments, considering different lighting and surface conditions. Overall, this system provides an accurate stair parameter extraction method that improves the stair climbing ability of people with vision impairments who use orthoses and prosthetic devices.