Parking Slot Detection and Maintenance Using Deep Learning

Abstract

Traffic is a major issue in Bangladesh, particularly in urban regions like Dhaka. The main issue with traffic in Bangladesh is parking by the wayside. Private transportation users spend the majority of their waking hours looking for parking spaces. I therefore made the decision to create a system that aids in the discovery of available parking spaces. In large cities, traffic and energy waste can be significantly reduced by the use of parking-management systems, including services that identify empty spaces. Since they may make use of the cameras that are already present in many parking lots, visual methods for spotting open places offer an affordable alternative. I created a system that is economical using a deep learning model. I have utilised three distinct deep-learning models. The three of them are Yolov7, Yolov8, and Mask-RCNN. Box loss, Class loss, Instances, Object loss, and mAp are the only outputs that the model produces. Yolov8 gives me the mAP of the models, at 96.8%. I have chosen the accuracy of YOLO v8 here because it is updated among all versions and gives very good and perfect accuracy. Here I have used the video footage of my university garage as a dataset. I have tried to give my best performance in every step of this thesis to use my own dataset. I'll try to make it so that authorised users can access the parking garage in the future. In order to achieve higher accuracy, and will expand the dataset.