基于激光雷达回波信号的雾天车道线快速检测  

Rapid detection of foggy lane markings based on LiDAR echo signals

作　　者：陈琼[1] 李小玲[1] CHEN Qiong;LI Xiaoling(Gongqing College of Nanchang University,Jiujiang 332020,China)

机构地区：[1]南昌大学共青学院,江西九江332020

出　　处：《中国测试》2024年第11期120-128,共9页China Measurement & Test

基　　金：2022年江西省教育厅科学技术研究项目(GJJ2203812);2022年江西省教育厅科学技术研究项目(GJJ2203801)。

摘　　要：当前主要通过深度神经网络模型提取路面车道线,并设计能见度检测网络,根据车道线可见长度检测路面车道线。但是,在雾天,基于深度分割神经网络设计编码解码结构非相关因素过多,无法通过其提取车道线特征图,无法准确检测图像坐标系下可见车道线的高度。针对雾天驾驶时的视觉障碍问题,以激光雷达技术为支撑,提出雾天车道线快速检测方法。根据激光雷达回波信号中每个回波脉冲宽度级的扫描点数,采用最小类内方差算法,阈值分割路面与车道线扫描点,由3σ准则分离出车道线的种子点后,基于高斯核函数加权搜索的生长准则,经区域生长得到完整的车道线种子点集。基于密度的空间聚类算法二次聚类获取的车道线种子点集,得到车道线的识别结果。以识别结果为基础,建立抛物线模型,结合随机采样一致性算法和最小二乘法,依据拟合分值迭代取得最优模型,通过拟合完成车道线检测。实验结果表明:该方法屏蔽雾天干扰引起的非相关因素,清晰检测出雾天环境中的多种车道线。在雾天环境车道线检测中,交并比高于0.95,F1值高于96%,可以满足准确性和实时性需求,为雾天驾驶提供有效的解决方案。Currently,deep neural network models are mainly used to extract road lane lines,and a visibility detection network is designed to detect road lane lines based on their visible length.However,in foggy weather,there are too many irrelevant factors in designing encoding and decoding structures based on deep segmentation neural networks,making it impossible to extract lane line feature maps and accurately detect the height of visible lane lines in the image coordinate system.A fast detection method for lane markings in foggy driving is proposed,supported by LiDAR technology,to address the issue of visual impairment during foggy driving.Based on the scanning points of each echo pulse width level in the LiDAR echo signal,the minimum intra class variance algorithm is used to segment the scanning points of the road surface and lane lines using a threshold.The scanning points are divided by 3σ.After separating the seed points of lane lines based on the criterion,a complete set of lane line seed points is obtained through regional growth using a growth criterion weighted by Gaussian kernel function search.The density based spatial clustering algorithm obtains the seed point set of lane lines through secondary clustering,and obtains the recognition results of lane lines.Based on the recognition results,establish a parabolic model,combine random sampling consistency algorithm and least squares method,iteratively obtain the optimal model according to the fitting score,and complete lane detection through fitting.The experimental results show that this method shields irrelevant factors caused by foggy interference and clearly detects multiple lane lines in foggy environments.In foggy environment lane detection,the intersection to intersection ratio is higher than 0.95 and the F1 value is higher than 96%,which can meet the accuracy and real-time requirements and provide an effective solution for foggy driving.

关 键 词：雾天环境 激光雷达 回波信号脉冲宽度 基于密度的空间聚类算法 抛物线 车道线检测 

分 类 号：TB9[一般工业技术—计量学] TN95[机械工程—测试计量技术及仪器]