Pothole.io

Final Soldered Product
Live Data Collection (Road Test)

Inspiration

We were frustrated with the state of the roads around Purdue and West Lafayette in general. Potholes are a common but often ignored problem. City's must hire people to go out and inspect for potholes, in a slow and expensive process. We strove to develop a passive system automatically detecting and mapping potholes with average vehicles.

What it does

Our device screens accelerometer and gyroscope data using machine learning to filter for false positives and detect potholes. After detection, the GPS data will be sent wirelessly through Wi-Fi to be processed in the Raspberry Pi. Here, we map the data to a google maps analog so that users may interface with known pothole locations.

How we built it

The final prototype was assembled through hands-on soldering, ensuring all components were properly connected for reliable functionality. We used a breadboard for rapid testing and experimentation, allowing us to quickly iterate and troubleshoot the circuit before finalizing the design. For the software side, we developed the User Interface (UI) using HTML, making it intuitive and user-friendly. Additionally, the machine learning component for pothole detection was trained and optimized to accurately identify road hazards. Throughout the process, collaboration and iterative testing played a key role in refining both the hardware and software components of our project

Challenges we ran into

We were impeded by faulty chips and inconsistent/incomplete sensor readings, and also faced limitations with Isolation Forest, the MLM we chose. This required careful calibration to achieve reasonable false-positive rejection.

Accomplishments that we're proud of

We were able to consolidate both sensors down to one ESP32, saving much space and development time, while still maintaining reliable performance. Additionally, we were able to in

What we learned

We learned to effectively interface with AI in helping to speed up software development. We also learned to troubleshoot within the limitations of the hardware we had access to and the tight deadline.