Smart Parking System

Overview

A team embedded-systems project pairing a Raspberry Pi 4 for vision-based occupancy detection with an Arduino Due for hard-real-time sensor polling, servo gate actuation, and driver-facing display feedback. This was a team project; I was responsible for the complete electrical and software system.

Team Project & My Contribution

Smart Parking System was developed as a team project. I was responsible for the complete electrical and software system, including its architecture, implementation, and testing.

• Electrical architecture and integration — sensor wiring, voltage-divider signal conditioning for the ultrasonic echo lines, LED current-limiting, and power distribution across the Raspberry Pi, Arduino Due, and sensor rail.
• Embedded software for both controllers — the Arduino Due's real-time sensor polling and actuator control, and the Raspberry Pi's coordination and decision logic.
• OpenMV camera integration for ROI-based parking-space occupancy detection.
• Device communication — UART handshaking between the Raspberry Pi, Arduino Due, and OpenMV camera.
• Control logic for the dual-sensor gate-timing and occupancy behavior, and testing/debugging of the full electrical and software system.

System Architecture

Testing & Results

• Gate actuation + ultrasonic sensing (hard real-time): ~600–900 ms gate response at 100 ± 2 ms polling — within the required under-1-second window.
• IR fail-safe layer: ~100 ms polling, used to confirm occupancy when the camera's read is uncertain.
• LCD display updates: ~500 ms, matching the design target.
• OpenMV detection: ran at 30–31 FPS against a ~30 FPS target.
• Field validation: tested against a real outdoor parking lot rather than only a lab bench, to confirm the ROI-comparison approach held up outside controlled conditions.

Challenges & Limitations

• Vision-based detection was sensitive to strong sunlight reflections and shadows during outdoor testing — the IR sensors were added as an override layer so occupancy detection stayed reliable when lighting interfered with the camera.
• A live camera-feed demo wasn't possible at presentation time (no OpenMV IDE connector on hand), so detection was validated against recorded real-world frames instead.