Related: https://github.com/sharare90/Smart-Walker
This project has two approaches. See DepthImages for obstacle recognition using depth images. See PointCloud for obstacle detection using point clouds.
Segmentation contains code from obstacle recognition using linear time-series segmentation.