An Adaptive Vision-based Outdoor Car Parking Lot Monitoring System

Monitoring system installed at a parking field is to update real time data of vacant parking lots to central management and to provide search and book services to car drivers. Camera attached with image processing unit running Deep Learning based algorithms to detect vacant/occupied parking lots is promising with rather high accuracy. In this work, we propose a novel solution using mAlexNet, a CNN-based model, to classify vacant or occupied state of each parking lot snapshot, with a pre-processing stage, camera adjustment method, enabling the solution to be automatically adaptive with the condition setting variations. The solution is capable to run on resource constrained processors like Rasberry Pi 4 and has been tested on parking fields at our university campus, showing the accuracy of over 97% and rather fast processing pace of 0.743 seconds in average for each frame capturing 24 parking lots.