Real-Time Parking Occupancy Detection using Motion Detection and Object Detection on Edge Device

abstract-The growing number of vehicles in urban cities escalating issue on limited parking spaces. Finding unoccupied parking spots for the drivers can be a challenge for them. Several studies have been conducted on the parking management system to provide information about parking occupancy in real time. However, the proposed methods still have flaws, most of them utilize sensors that need installation that require extra cost. A computer vision with object detection can mitigate this issue with only a single deployment by using lightweight edge device. Nevertheless, the majority of object detection algorithms need a lot of processing power usage, so another concern is power efficiency. In this paper, we construct a real-time parking occupancy detection system by combining motion detection and object detection on edge device. We utilize OpenCV library to perform motion detection by comparing each frame to detect any motion. For object detection, we collected a dataset sample of parking space in campus and trained to create a new model specified for parking occupancy detection. The result shows that our trained model object detection based on YOLOv8n performed with average accuracy of 99.86% for our parking space scenario with performance over 5 FPS on edge device. Also combining motion detection, it can reduce CPU usage by 44.78% and watt-hour usage by 52.17% on our edge device.