Design and Simulation of Smart Parking System Using Image Segmentation and CNN

Abstract

Reducing the congestion of busy parking lots by giving people in the vicinity an accurate idea of how many spots are open is a feature that smart parking systems can provide. so far, these systems have been deployed mostly for indoor locations using expensive sensor-based technology. As research and development of image-based detection techniques has increased, it follows that many commercial products are using smart parking technology, and thus, the need for the systems is growing. This research uses a binary Support Vector Machine (SVM) classifier with an image classifier trained using a Convolutional Neural Network (CNN) to identify the presence of vehicles in parking spots. Classifier training and testing used deep CNN features drawn from public datasets with varying light and weather conditions. So, we check how well the technique does with regard to transfer learning using a dataset designed for our study. We've concluded that our approach is good for solving issues in outdoor settings, as shown by our 99.7 percent detection accuracy and 96.7 percent accuracy for the public dataset and our dataset, respectively.