A dashcam-based computer vision system that watches for parking enforcement vehicles from a parked car and sends a reminder to top up before the session lapses. It solves a mundane problem: losing track of a parking session while heads-down at work, and finding out it expired from the ticket on the windshield.

The system runs entirely on a Raspberry Pi 4B with a Camera v3, which sets the real engineering constraint. YOLOv8 inference on a continuous video stream, on a board with no GPU, has to keep a frame rate high enough to catch a vehicle passing at street speed. OpenCV handles frame preprocessing to make flashing lights separable from ordinary traffic, headlights, and reflections, and the detection model was trained in PyTorch on a hand-annotated dataset built with LabelImg from footage captured across varied lighting and weather conditions. Identifying an enforcement vehicle is not a single-frame problem, since the distinguishing signal is a temporal flash pattern rather than a static shape.

A GPS module tags each detection, building a location-stamped log of enforcement presence over time so the system can learn patrol frequency by area rather than reacting only to what is currently in frame. Detections push an alert to the user, and the timing logic is deliberately advisory: it prompts a decision rather than acting on the user's behalf.

Status: In progress. Detection, GPS logging, and alerting are working. Automated payment through a parking provider API was scoped but not implemented, so the system currently reminds rather than pays.