OpenWalk

OpenWalk is an open-source pedestrian walkability perception stack that brings autonomous-driving-style environmental sensing to human-scale navigation. Using first-person video, it detects sidewalk hazards such as ice, snow, construction barriers, curbs, missing tactile paving, and uneven surfaces, then converts them into proactive audio or haptic alerts tailored to different mobility profiles. The project targets blind and low-vision pedestrians, wheelchair users, elderly pedestrians, and others who need ground-level safety information that current maps cannot provide. Its core output is a reusable open-source research foundation: dataset schema, annotation guidelines, baseline models, scoring rules, benchmark protocol, and privacy-preserving data collection tools.