Lane Line Detection in Parking Lot Scenarios

With the widespread adoption of electric vehicles, intelligent charging robots in parking lot environments must address efficient navigation in complex spaces, which relies fundamentally on lane line detection technology. Traditional lane detection methods depend on handcrafted feature extraction and suffer from insufficient robustness. While deep learning-based approaches have made progress, they exhibit limitations such as poor real-time performance, redundant information, and weak adaptability to diverse scenarios<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">]</sup>.To tackle challenges like occlusion, wear, and multidirectional layouts of parking lot lane markings, this paper proposes PLNet, a neural network model that integrates multiscale information within an object detection framework to enhance detection accuracy. Innovatively, we introduce a LineIOU loss function to optimize detection consistency. Experiments demonstrate that PLNet achieves a balance between accuracy and speed across backbone networks like ResNet and MobileNet:With ResNet18, PLNet achieves an F1@50 of 81.2 at 196 Frames Per Second(FPS), outperforming LaneATT by 10.3% in accuracy under the same network configuration.MobileNet maintains 78.5 F1@50 at 223 FPS, surpassing baseline models by 1.05% in overall performance metrics.The model achieves a peak accuracy of 81.8 F1@50 (ResNet101) while retaining real-time performance (>196 FPS). All variants of PLNet consistently exceed 78 F1@50, significantly resolving low-visibility lane line issues in complex environments.This study confirms that PLNet provides a lightweight, high-precision perception solution for intelligent charging robots. Its efficiency-accuracy trade-off ratio—4.14% speed gain per 1% accuracy drop—highlights practical engineering value, advancing the deployment of flexible charging infrastructure.