Combining deep learning methods and rule-based systems for automatic parking space detection

This paper presents an Automatic Parking Space Detection (APSD) algorithm designed to reduce traffic in cities while offering an information system of available parking zones. The main aim of such a system lies in its ability to identify parking spaces in a distributed manner, achieved by installing multiple APSD systems across a fleet of vehicles. This fleet, during its regular operations, communicates the availability of parking spaces to a centralized information system. Our methodology employs a rule-based system that seamlessly integrates a variety of neural networks for different specific tasks. These tasks include depth estimation, road segmentation, and vehicle detection. This approach would fall into a modular category instead of an end-to-end solution, using the M & aacute;laga Urban Dataset in the experiments. We present a preliminary experiment for parameter settings and an ablation study to quantify each subsystem contribution to the results. The proposed system achieves a parking space detection F1 score of 0.726.