Optimizing YOLOv8 for Parking Space Detection: Comparative Analysis of Custom YOLOv8 Architecture
Apar Pokhrel, Gia Dao
TL;DR
This paper evaluates how integrating different backbones into YOLOv8 affects parking space occupancy detection on the PKLot dataset. By comparing ResNet-18, EfficientNetV2-S, VGG-16, and Ghost-P2 backbones, the study highlights trade-offs between detection accuracy and computational efficiency. EfficientNetV2-S achieves the best overall performance with high precision and a favorable FLOP count, while ResNet-18 offers a strong balance between accuracy and speed, and Ghost-P2 favors edge deployment due to ultra-fast inference. The findings inform deployment decisions for intelligent parking management, including potential edge implementations and future work on deeper backbones and additional modalities.
Abstract
Parking space occupancy detection is a critical component in the development of intelligent parking management systems. Traditional object detection approaches, such as YOLOv8, provide fast and accurate vehicle detection across parking lots but can struggle with borderline cases, such as partially visible vehicles, small vehicles (e.g., motorcycles), and poor lighting conditions. In this work, we perform a comprehensive comparative analysis of customized backbone architectures integrated with YOLOv8. Specifically, we evaluate various backbones -- ResNet-18, VGG16, EfficientNetV2, Ghost -- on the PKLot dataset in terms of detection accuracy and computational efficiency. Experimental results highlight each architecture's strengths and trade-offs, providing insight into selecting suitable models for parking occupancy.