Proposta de Modelos Leves para Classificação de Vagas de Estacionamentos em Cidades Inteligentes

In smart cities, a common problem is the parking spots classificationinto empty and occupied. It may seem simple, but a large numberof Deep Learning approaches rely on CNNs (Convolutional NeuralNetworks). These solutions are commonly expensive, demandinghigh computational power and specialized hardware to run properly,making them unsuitable for large-scale deployments, such asin smart cities. In this work, we propose two lightweight CNN architectures,built upon existing solutions by enhancing their efficiencyand robustness. We used a cross-dataset scenario, where a modelis trained and validated in two datasets and tested in another, applyingthree robust state-of-the-art datasets: PKLot, CNRPark-EXTand PLds. This process improves generalization across differentcontexts and sets a more realistic scenario when compared to realurban environments. Also, we compared our models to state-ofthe-art networks, such as MobileNetV3 Large and Small to ensureconsistency and validate the results with well-explored modelsin the literature. Our results showed that our models, with up to34× and 88× fewer parameters than the MobileNetV3 Large, reachless than 2% lower accuracy when compared to the MobileNetV3networks. Furthermore, by using grayscale images, the results wereslightly better and also decreased processing and storage costs.