Deep learning-based parking occupancy detection framework using ResNet and VGG-16

The rise in traffic congestion today has necessitated growing research and development in parking management systems to provide real-time indications of the occupancy of indoor and outdoor parking spaces. The main challenge has been developing affordable detection methods based on images to substitute the more expensive sensor-based techniques deployed in indoor environments. With the advancement in computer vision and deep learning, we aim to harness the remarkable performance of convolutional neural networks for carrying out image category recognition tasks to develop a robust parking occupancy detection framework. The classifier was modeled and evaluated with the help of the features learned by the model from the PKLot dataset under varied illuminance and weather conditions. These two models used for parking space detection and classification include – Resnet50 (combined with support vector machine) and VGG16 (combined with OpenCV functionalities). An accuracy of 98.9% was reported with the ResNet and support vector machine model. The model using VGG16 reported an accuracy of 93.4%. Thereby, a low-cost and reliable solution to parking occupancy systems in outdoor systems was arrived at.