Development of an Automatic Data-centered Parking Space Occupancy Detection with Vehicle License Plate Web-Based Monitoring

Authors: Abdul Ghaffar U. Tagolanao, Hashim Monir M. Ambia, and Cristina P. Dadula

Abstract

This paper is a two-step camera-based system running on a web-based monitoring center that utilizes Convolutional Neural Networks (CNN) to achieve the detection of license plates and the identification of the vacancy of parking spaces for a display to entering drivers. The license plate detection is achieved using a retrained tiny You Only Look Once (YOLO) model in conjunction with Tesseract OCR for character recognition. The detected plate images are stored in separate directories according to date, and filenames according to time. The parking occupancy network employs a simplified AlexNet to classify parking spaces as either ’vacant’ or ’occupied’. Testing was done on an emulated parking lot, where the live camera feed on the entrance and exit are simulated by prerecorded videos. The license plate detection works on an average of 0.023 seconds with 100% accuracy on still images. Parking Space occupancy identifier can classify the vacancy of 37 parking slots in an average of 0.06 seconds with 10.14% of false occupancy identification, 0.92% of false vacancy identification, and 88.94% correct identification. Though tested in a simulated environment, the test results show that the system can be implemented and applied to actual parking lots.