顾及空间特征的杆目标渐进式识别分类方法  

A Progressive Recognition and Classification Method for Pole Targets Considering Spatial Features

作　　者：田茂义[1] 张佳琦 刘如飞[1] 张振虎 李子豪 Tian Maoyi;Zhang Jiaqi;Liu Rufei;Zhang Zhenhu;Li Zihao(College of Geodesy and Geomatics,Shandong University of Science and Technology,Qingdao 266590,Shandong,China;Shandong Expressway Engineering Inspection Co.,Ltd.,Jinan 250002,Shandong,China;Innovation Research Institute of Shandong Expressway Group Limited,Jinan 250098,Shandong,China)

机构地区：[1]山东科技大学测绘与空间信息学院,山东青岛266590 [2]山东高速工程检测有限公司,山东济南250002 [3]山东高速集团有限公司创新研究院,山东济南250098

出　　处：《应用激光》2025年第2期132-140,共9页Applied Laser

基　　金：国家自然科学基金资助项目(42106180);山东省科技型中小企业创新能力提升工程项目(2022TSGC1135);“菁英计划”科研支持项目(0104060541613);2023年山东省交通科技创新计划项目(2023B78-01)。

摘　　要：针对车载激光点云部分缺失导致的杆柱漏识别、杆顶错分类的问题,提出一种顾及空间特征的杆目标渐进式识别方法。该方法利用级联随机森林模型以及空间关系特征,有效地解决了点云中断续杆目标的准确识别和分类问题。首先,通过级联结构的上层模型在多尺度的节点切片中的圆弧形态特征获取垂直方向连续圆弧间的空间关系特征,结合圆弧形态特征实现断续杆柱的精确识别;其次,基于级联模型的杆柱识别结果,获取杆柱垂直簇的顶部聚类体,并通过空间关系约束的ESF形状特征获取杆柱与杆顶的空间特征,结合PCA维度特征利用级联模型实现杆顶准确分类。实验结果表明,该方法在两组实验数据中的识别准确率分别达到了96.56%和94.51%,能够有效应对复杂道路场景中杆目标的识别与分类问题,且具有较强的稳定性。Aiming at the problems of missing identification of pole columns and misclassification of pole tops caused by partial missing of vehicle-mounted laser point cloud,a progressive pole target identification method considering spatial features is proposed.The method effectively solves the problem of accurate identification and classification of intermittent pole targets in the point cloud by using a cascaded random forest model and spatial relationship features.Firstly,the spatial relation features between continuous arcs in the vertical direction are obtained by the arc morphology features of the top model of the cascade structure in the multi-scale node slices,and the accurate identification of intermittent poles is achieved by combining the arc morphology features;secondly,based on the pole identification results of the cascade model,the top clusters of the vertical pole clusters are obtained,and the spatial features of poles and pole tops are obtained by the spatially relationally constrained ESF shape features combined with PCA dimensional features using the cascade model to achieve accurate classification of pole tops.The experimental results show that the recognition accuracy of this method reaches 96.56% and 94.51% in two sets of experimental data,which can effectively deal with the recognition and classification of pole targets in complex road scenes and has strong stability.

关 键 词：杆目标 级联随机森林 多尺度切片 圆弧 空间特征 

分 类 号：TN249[电子电信—物理电子学]