PathAble

Inspiration

Indoor routes can change when a chair or other object blocks a walkway. We wanted to explore how a fixed camera could help report an obstacle and share that information with the people responsible for reviewing the route.

What it does

PathAble uses a fixed camera assigned by configuration to junction B and edge B–C. OpenCV 5 and YOLO11s segmentation help detect a chair overlapping the camera’s defined region across multiple frames. The camera client sends a photo and report to the AWS backend.

A reviewer checks the photo. If they approve the report, PathAble marks B–C as blocked on the shared map. The laptop receives the updated map, proposes a new route, and announces it aloud. In our demo, approval updated the map to version 4 and the laptop announced B → E → F → D.

PathAble does not measure distance or corridor clearance. A proposed route is not proof that it is safe to walk.

How we built it

We split the system into three parts:

Camera client: Python, OpenCV 5, and YOLO11s segmentation detect objects in a configured camera region and submit reports with photos.
AWS backend: API Gateway and Cognito handle the API and sign-in. AWS Lambda runs the backend; DynamoDB stores report and map data, and S3 stores private photo evidence.
Reviewer and laptop clients: The reviewer dashboard displays pending reports and sends an approval or rejection. The laptop client receives map changes, calculates a proposed route on its local map, and uses Windows speech to announce it. The camera location and edge are assigned through configuration. Distance and clearance remain Unknown because this prototype does not measure them. ## Challenges we ran into A camera detecting an object in an image does not tell us how far away it is or how much usable space remains. We had to keep detection separate from human approval and avoid presenting unknown measurements as facts.

We also had to connect the camera, reviewer, AWS backend, and laptop so that one approved map change reached the navigation client. Capturing café images raised privacy concerns, so we used a view without customers or private details for the AWS run.

Accomplishments that we're proud of

We completed a live camera-to-AWS review flow: the camera sent a report and photo, a reviewer approved it, AWS returned an updated shared map, and the laptop received the update and spoke the proposed detour.

The approval changed the map to version 4 and marked B–C blocked. The reviewer remained in control of the map change.

What we learned

Computer vision can help flag a possible obstacle, but an image alone cannot confirm physical clearance or a safe route. Human review, clear uncertainty, and careful separation between a proposed detour and verified walking guidance are essential.

What's next for PathAble

Before PathAble could guide real users, we would need to measure and verify corridor clearance, check alternate routes in a controlled environment, and evaluate the experience with users who have mobility needs. We would also test offline behavior, retries, rejected reports, and storage recovery. Until then, PathAble is a prototype for obstacle reporting and human-reviewed map updates—not a system that certifies a route as safe.