Inspiration
We were inspired by the idea of making spatial awareness more accessible using AI. Busy environments can be overwhelming, and we wanted to explore how computer vision could translate visual information into simple, meaningful audio cues. Instead of building something flashy, we focused on solving a real problem in a practical way. The goal was to create a lightweight assistive system that works anywhere, even without internet.
What it does
VisionHat is a real-time assistive vision system that detects objects in front of the user and provides audio warnings when something is getting too close. It runs entirely offline on a Raspberry Pi using YOLOv8 for object detection. The system prioritizes objects directly in the user’s path and calculates proximity based on bounding box size and position. When an object enters a defined danger zone, it announces a clear warning through Bluetooth audio.
How we built it
We built the system using a Raspberry Pi, a camera module, and YOLOv8 running locally for object detection. We processed live video frames, filtered specific object classes, and designed a proximity logic system based on relative bounding box area growth. Audio feedback was implemented using text-to-speech routed through Bluetooth headphones. To ensure smooth performance, we optimized frame resolution, reduced unnecessary rendering, and fine-tuned detection thresholds.
Challenges we ran into
One of our biggest challenges was performance optimization on edge hardware. Running real-time computer vision on a Raspberry Pi required careful tuning to balance speed and accuracy. We also faced Bluetooth audio configuration issues, camera indexing problems, and offline deployment constraints. Debugging without relying on cloud services pushed us to better understand system-level configuration and resource management.
Accomplishments that we're proud of
We are proud that the entire system runs fully offline on affordable hardware. Achieving stable real-time object detection with audio feedback on a Raspberry Pi was a major milestone. We successfully integrated computer vision, proximity logic, and Bluetooth audio into a cohesive, working system. Most importantly, we built something that feels practical and demo-ready rather than theoretical.
What we learned
We learned how to deploy AI at the edge and how important optimization is when working with constrained hardware. We gained experience in real-time inference, system debugging, and hardware-software integration. This project also reinforced the importance of designing for clarity and usability instead of complexity. Building something that works reliably offline taught us a lot about robustness and system design.
What's next for dondstrez_VisionHat
Next, we want to improve performance and expand object awareness with smarter path prediction and obstacle tracking. We are also exploring wearable integration to make the system more compact and comfortable. Future versions could include haptic feedback and improved depth estimation for more accurate proximity detection. Ultimately, we want to refine it into a polished, portable assistive device that can operate reliably in real-world environments.
Log in or sign up for Devpost to join the conversation.