Inspiration
Getting lost in the wilderness is a terrifying reality when GPS signals fail under thick forest canopies or in deep mountain canyons. We realized that while modern hikers rely heavily on digital maps and satellites, nature often blocks those very signals. We wanted to build a safety net that does not depend on a cloud connection or a satellite orbit. Our goal was to create a survival tool that works anywhere, even in the most isolated environments on Earth, by using the one thing that never loses signal: the visual landscape.
What it does
TrailSense is an edge AI navigation system that uses visual memory to guide users back to safety. As a hiker moves away from their starting point, the system captures digital breadcrumbs of the terrain. It stores these unique landmarks in a local database on the device. When it is time to return, TrailSense compares the live camera view against its stored visual memory to confirm the path. It provides real time guidance through hardware indicators, ensuring that a user can retrace their exact steps without ever needing a cellular bars or a GPS lock.
How we built it
We utilized the Arduino Uno Q as the brain of the project, taking advantage of its powerful Qualcomm processor to handle complex vision tasks. An ESP32 camera acts as the eyes of the system, capturing images of the trail. The Linux side of the board runs a Python environment where OpenCV logic calculates the similarity between the current surroundings and saved landmarks. We used the internal bridge to link this high level analysis to the microcontroller side, allowing us to trigger physical alerts and status lights with zero latency.
Challenges we ran into
One of the biggest hurdles was configuring the computer vision environment on a new hardware architecture. We spent significant time resolving library dependencies and ensuring that the Python environment could interact smoothly with the ARM64 processor. Another major technical wall was the synchronization between the two processors. We had to ensure that the visual data being processed on the Linux side could communicate its results back to the hardware side quickly enough to provide meaningful navigation cues without the system timing out or crashing.
Accomplishments that we're proud of
We are incredibly proud of achieving reliable visual landmark recognition entirely at the edge. Seeing the system correctly identify a previously seen rock formation or tree line without any internet access was a huge milestone. We successfully bridged the gap between complex software logic and physical hardware, creating a tool that feels both intelligent and rugged enough for the outdoors.
What we learned
This project taught us the immense potential of dual processor embedded systems for safety applications. We gained deep insights into how to optimize vision algorithms for performance on mobile processors and learned how to manage local storage for data intensive tasks like image logging. We also discovered that feature based matching is an incredibly robust way to handle navigation in shifting outdoor lighting conditions.
What's next for TrailSense
The future of TrailSense involves making the guidance even more intuitive. We plan to integrate local generative AI to provide descriptive analysis of landmarks, such as identifying specific plant species or geological features. We also want to explore adding haptic feedback into gear like backpack straps, allowing for a completely eyes free navigation experience that lets the hiker focus on the beauty of the trail.
Log in or sign up for Devpost to join the conversation.