IRCHAD — Indoor Navigation for the Visually Impaired

Finding your way inside a large building is a daily obstacle for people with a visual impairment. IRCHAD was our fourth-year project at ESI: an indoor navigation and assistance system for visually impaired users, with apps for the people around them too.

We were a team of twelve, and I led it. We split the system into Android apps for users, caregivers and maintenance technicians, a web platform, and backend services: twelve repositories in all. By the end, live indoor navigation worked end to end, through an embedded device and the user's mobile app.

What I built

An indoor map starts as an architectural floor plan, which is just an image. I built the preprocessing pipeline for those plans, trained a YOLO model on Kaggle to detect doors, walls and windows, and designed the protocol to evaluate it; it reached 89% mAP. I then served the detector through a FastAPI endpoint that returns its detections as GeoJSON.

On the platform side, I built the web app's dashboards for decision-makers, its sales module and its authentication, and the sales API of the statistics service.