稀疏点云道路分割与低矮路障检测的自适应方法  

Adaptive Road Segmentation and Low Obstacle Detection on Sparse LiDAR Point Cloud

作　　者：罗俊奇 叶勤 张绍明 史鹏程 LUO Junqi;YE Qin;ZHANG Shaoming;SHI Pengcheng(College of Surveying and Geo-information,Tongji University,Shanghai 200092,China;School of Computer Science,Wuhan University,Wuhan 430072,China)

机构地区：[1]同济大学测绘与地理信息学院,上海200092 [2]武汉大学计算机学院,武汉430072

出　　处：《遥感信息》2022年第6期108-115,共8页Remote Sensing Information

基　　金：国家自然科学基金项目(41771480);上海市自然科学基金项目(22ZR1465700)。

摘　　要：针对稀疏点云道路分割与路障检测,特别是低矮路障检测方法中存在严重阈值依赖,从而受限于特定应用场景的问题,提出了一种自适应阈值的道路分割与低矮路障检测方法。通过坡度自适应算法与局部平面拟合,改进LineFit道路地面提取算法以实现斜率阈值自适应。利用基于曲率突变的点云分割算法实现道路边界提取。在道路分割的基础上,引入目标点云的局部相对密度实现了欧式聚类低矮路障检测的半径阈值自适应,提高了不同道路场景低矮路障检测的鲁棒性。实验表明,与传统的点云道路分割与路障检测方法相比,该方法对稀疏点云的道路分割与路障检测精度达到90%,对低矮路障具备更高的检测召回率与精确率,适用于低线束LiDAR自动驾驶平台的道路环境感知。For the problem that the fixed thresholds are necessary for existing point cloud segmentation methods used in different road scenes,an adaptive road segmentation and low obstacle detection method on sparse LiDAR point cloud is proposed.Firstly,the proposed method improves the LINEFIT ground segmentation algorithm with a slope threshold calculation way.Then,based on curvature mutation,a road edge extraction method is proposed.Finally,an adaptive low obstacle detection method is proposed and the local relative density is used to calculate the radius threshold in Euclidean clustering algorithm,which avoids the influence of different point cloud density on the detection.Experimental results show that the accuracy of proposed method for road segmentation and obstacle detection reaches 90%,and the low obstacle detection shows better than traditional Euclidean clustering algorithm.Proposed method can be well applied to the road environment perception of low harness LiDAR autonomous driving mobile platform.

关 键 词：稀疏点云 低矮路障 阈值自适应 点云分割 目标检测 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]