A Camera-based Smart Parking System Employing Low-complexity Deep Learning for Outdoor Environments

Abstract

The smart parking occupancy detection system is a technology which aims to mitigate the traffic congestion problems by reducing time for drivers to look for vacancy positions in car parking lots and providing efficient parking space utilization. Several reports have shown that the smart parking system not only alleviates traffic problems but also drives business growth and economic development within that neighborhood. In this paper, we propose a computer vision-based smart parking lot occupancy detection system employing low-complexity deep neural network architecture. A smart camera system which consists of a Raspberry Pi 3 attached to a camera utilizes a reduced-complexity deep neural network model to detect vacancy positions. We train and cross-validate our model using PKLot-Val dataset and test the performance of our model using PKLot-Test and SWUpark datasets. SWUpark dataset has been created in the context of this research, accumulating visual information of parking lots at Srinakharinwirot University across several weather conditions. Through exhaustive hyperparameter tuning and stochastic gradient descent optimization, our model achieves 88% accuracy, almost 15% higher than those obtained from state-of-the-art approach.