复杂变道场景下的轻量化车道线检测算法研究

Research on Lightweight Lane Detection Algorithm in Complex Lane Changing Scenes

设计了一种轻量化的车道线检测算法用来解决复杂变道场景下存在遮挡并可能伴随阴影、光线不足等不利因素时的车道线识别任务。首先对主流的车道线检测算法逻辑作简要概述,进一步剖析其存在的不足。随后从轻量化的主干特征提取网络RepVGG-A0、路径聚合网络模块、先验车道线特征提取模块、损失函数设计等方面展开进行介绍,从整体上搭建了基于跨层特征融合的轻量化车道线检测网络。最后利用CULane数据集进行相关测试并完成了与其他相关检测算法的性能对比。测试结果表明,提出的轻量化车道线检测算法能够在保持检测精度良好的情况下速度达到132帧每秒,在经TensorRT推理加速后,检测速度突破220帧每秒,充分达到了实时性检测的要求。

A lightweight lane detection algorithm is designed to solve the task of lane recognition in complex lane change scenes when there are obstacles and may be accompanied by shadows,insufficient light and other adverse factors.Firstly,the logic of the mainstream lane detection algorithm is briefly summarized,and its shortcomings are further analyzed.Then,from the aspects of lightweight backbone feature extraction network RepVGG-Ao,path aggregation network module,prior lane feature extraction module,loss function design,etc.,the lightweight lane detection network based on cross-layer feature fusion is built as a whole.Finally,CULane data set is used for correlation testing and performance comparison with other correlation detection algorithms is completed.The test results show that the proposed lightweight lane detection algorithm can reach 132FPS while maintaining good detection accuracy.After accelerated by TensorRT reasoning,the detection speed exceeds 220FPS,which fully meets the requirement of real-time detection.