Detection and counting of vehicles using deep learning in a parking lot area of a university institution

Franco Gonzales del Valle Romero G.; Jesús Neyra Espinoza W.; Huamaní-Navarrete P. · Proceedings of the LACCEI international Multi-conference for Engineering, Education and Technology (2024) · DOI: 10.18687/laccei2024.1.1.1772

Resumen

This article describes the implementation of two artificial neural networks with deep learning, with the objective of detecting and counting vehicles in the main parking area of a university institution in Lima-Peru, Universidad Ricardo Palma. In this way, it became possible to know the number of available parking spaces to reduce consultation time by the security personnel of said institution. Convolutional neural networks were used: You Only Look Once v5 (YOLO v5) and Faster Region-based Convolutional Neural Network (Faster R-CNN), and Transfer Learning was applied. The dataset was made up of images from the place and obtained from the web. To count vehicles, space limitation was used, by the line of interest (LOI) and region of interest (ROI). And, for the graphical user interface, the Tkinter library was chosen, which allowed numerical visualization of the detected vehicles and available spaces. Finally, for real-time implementation, a mobile phone was installed above the third gate of the university itself, using the university's Internet network and connected to a laptop; and, when performing the analysis for the two thresholds used, 0.35 and 0.55, and 27 parking spaces, the YOLO v5 neural network gave a lower average error equal to 5.33%, while the Faster R-CNN neural network did so with 7.82%; On the other hand, the same YOLO v5 network reached a higher average precision value, 81.62%, in training with 100 epochs.

Instituciones: Universidad Ricardo Palma