Система автоматизованого моніторингу стану зони паркування автомобілів із використанням згорткових нейронних мереж

Abstract

This work presents the design and study of an automated system for monitoring the state of a car parking zone using computer-vision methods and convolutional neural networks. The system acquires video data from surveillance cameras, detects vehicles in the frame, localizes and recognizes vehicle licence plates, and determines the occupancy of parking spaces in real time. The YOLOv11 architecture is chosen for object detection, while occupancy control is implemented using oriented bounding boxes (OBB) and a polygonal representation of parking zones. Plate characters are recognized with Tesseract OCR after image pre-processing performed with the OpenCV library; the model is trained on the PyTorch framework. The system was validated by experimental testing on real footage of parking lots; the average accuracy of vehicle detection and occupancy determination reached 89 %.