One Convolutional Layer Model For Parking Occupancy Detection

Convolutional Neural Networks (CNN) have recently performed wonders in image recognition tasks. In this paper, we propose a new CNN model composed of one convolutional layer, which we called 1Conv. We apply 1Conv for the problem of parking space detection. We used the most popular datasets to evaluate the performance of our model that are: National Research Council Park (CNRPark), National Research Council Park Extension $(\mathbf{CNRPark}+\mathbf{EXT})$ , and Parking Lot (PKLot). We compared the results with mAlexNet, a CNN model similar to 1Conv. The results show that our model outperforms mAlexNet in terms of accuracy, Area Under the Curve (AUC), and execution time. The better accuracy of 1 Conv compared to mAlexNet was 99.06% against 90.71 % using CNRPark dataset. Which means that our model outperforms mAlexNet by 9% in term of accuracy. Execution time of mAlexNet is double compared to 1Conv.