Navi Grid

• 

  First Floor Annotations

• 

  Lower Level Annotations

Inspiration

We were inspired by the following research papers and GitHub repositories:

https://arxiv.org/abs/2512.12177 (Floorplan2Guide: LLM-Guided Floorplan Parsing for BLV Indoor Navigation)

https://github.com/chestnutforestlabo/Snap-and-Nav (Snap&Nav: Smartphone-based Indoor Navigation System for Blind People via Floor Map Analysis and Intersection Detection)

These works demonstrated how floorplans can be used for indoor navigation. We adopted some of the theoretical concepts from these projects and extended them by incorporating additional computer vision techniques to make the navigation process more seamless.

What it does

Navi Grid is an app that helps users navigate unfamiliar indoor spaces using just their smartphone camera. The system takes building floorplans as input and converts them into a navigation system that guides users through indoor environments.

How we built it

We used computer vision techniques such as Optical Character Recognition (OCR) to detect room numbers through the device’s camera. This allows the system to update the user’s location and adjust navigation guidance dynamically as the user moves through the building.

Challenges we ran into

We encountered several challenges while building this project:

1) We currently have to manually annotate the rooms on the floorplan. In the future, we plan to use Vision-Language Models (VLMs) to automate this process.

2) Due to time constraints during the hackathon, we were only able to experiment with OCR. In future iterations, we would like to explore additional computer vision techniques such as YOLO-based object detection and edge detection, which could help pinpoint the user's location more accurately.

3) We were working with limited processing power, which prevented us from using more advanced Vision-Language Models during the hackathon.

Accomplishments that we're proud of

Despite the challenges, we were able to build a functional prototype. In less than 24 hours (overnight into Sunday morning), we developed an indoor navigation system.

There were moments when some of us thought we might not be able to produce a working product, but through teamwork and persistence we were able to deliver a functioning system by the end of the hackathon.

What we learned

Since our project was inspired by academic research, we learned that while research papers provide strong theoretical foundations, they often work well only in specific scenarios.

To make these ideas work in real-world applications, it is necessary to adapt and iterate on them. Much of our hackathon time was spent going back and forth between theory and implementation to make the system work for our use case.

What's next for Navi Grid?

This hackathon was the first step for Navi Grid. We now know that it is possible to build an indoor navigation system using computer vision and Vision-Language Models.

In the future, we want to:

• Automate room annotation using VLMs
• Improve location detection with more advanced computer vision

Built With

• python