PSDNet: A Breakthrough Parking Space Detection Network Powered by YOLOv8

Parking congestion presents a significant challenge in urban areas worldwide, spanning both developed and developing nations. The escalating volume of vehicles compounds the scarcity of parking spots, particularly in downtown areas and adjacent streets. Local authorities are tasked with enforcing regulations to mitigate this issue, yet the current state of affairs often falls short of meeting the needs of numerous residents. Consequently, this article advocates using the YOLOv8 algorithm to identify vacant parking spaces. The proposed YOLOv8 comprises three pivotal blocks: the Backbone, Neck, and Head. The Backbone segment employs an enhanced version of the CSPDarknet53 network, integrating cross-stage partial connections to strategically strengthen information exchange across its diverse layers. The Neck section merges features with varying scales and incorporates contextual information from the feature maps. The Head section predicts class, objectness scores, and bounding boxes of identified objects. The performance of the YOLOv8 algorithm in identifying vacant parking spaces is evaluated using the parking space availability dataset. It achieves an F1-score of 98.7%, which is 8.7% higher than the existing object detector.