BeaverNav
Dec 2022
Problem
MIT is a large campus with thousands of rooms/hallways/buildings/staircases/elevators. Finding one’s way around campus can be difficult and daunting.
Our Solution
Indoor navigation for MIT’s campus: Enables users to navigate between rooms, provides routes from start destination to end destination
Note this is still a work in progress & is not yet a live website
We built this using the following pipeline:
- Scrape Floor Plans from MIT website, and process into images
- Extract key information (room coordinate locations & names) via pre-trained neural network text detection and recognition models (EAST, Easy OCR).
- Remove all doors & text from the floorplans & downsample (reduce resolution)
- Create a graph treating every pixel as a node in the graph
- Condense graph offline by running an additional all pairs shortest paths algorithm (APSP) & use A* for online planning
- Create an “abstracted graph” of MIT’s campus using elevators, staircases, entry-exits between buildings
More in-depth explanation can be found below