| dc.description.abstract |
Using sensors to detect the status of a parking space requires a large budget to install sensors in the parking space. Most of the parking spaces in densely populated areas have CCTVs for security reasons. The CCTV system can be further developed to detect the number of available parking slots and the location of vacant spaces. Our objective is to create a deep learning model for parking lot occupancy detection. This research used image datasets from PKLot and CNRPark. The data were divided into 2 sets with a ratio of 80:20, 1) a training dataset of 522,182 images, and 2) a test dataset of 130,519 images. The architectures chosen for modeling were Alexnet, VGG16, and RestNet50. The models’ performance was measured using the test dataset in order to select the best architecture to implement further. The best result is the Alexnet architecture achieved an accuracy of 99.20%, precision of 98.40%, recall of 98.60%, and an F1 score of 99.10%. The selected model was developed into a web application called ParkHere!, using React Js, Fast API, and PostgreSQL technologies to simulate the system of parking lot occupancy detection from CCTV images. |
|