This is my final project for Computer Vision.
This report presents a computer vision algorithm for parking lot occupancy detection. It is based on the use of YOLOv5, a one-stage deep learning object detector. The proposed solution is tested on CNRPark, a dataset containing images of the parking lot of the CNR (National Research Council) in Pisa. The images are taken on different days and times, from different viewpoints and with different light and weather conditions. Some of them include shadows and occlusions which make the occupancy detection task even more challenging. The results of the evaluation show that the algorithm is effective as long as its parameters are properly tuned. If so, the proposed approach proves to be robust not only to the variety of CNRPark but also to shadows and occlusions.
 |
 |
 |
 |
 |
 |
 |
 |
 |
