This week, the team made substantial progress toward a fully integrated end-to-end system. The most significant milestone achieved was completing the upload hardware-to-software-to-web-app pipeline: the dash cam now captures footage on the Raspberry Pi 5, transmits pothole detection metadata via the Blues Notecard over LTE, triggers a Cloud Function to receive and process the data, and displays the results live on the web app. This validates the core data flow of PotSpot across all three subsystems.
On the dash cam side, the team switched to the AI camera module as an interim solution, as the global shutter camera requires a lens to operate correctly. The software camera and inference pipeline were fixed, including a retrain of the YOLOv12n model for better detection accuracy, and the upload pipeline to the cloud backend was completed. Integration testing on real dash cam mock footage has also begun in preparation for the demo. On the web app, light and dark mode support was added, allowing users to select their display preference. The team is currently on schedule with the project timeline. Next steps include integrating live GPS polling into the detection pipeline and conducting in-vehicle testing by powering the Raspberry Pi 5 from the car’s cigarette lighter socket.