基于时空连续补偿的矿山可通行区域识别方法

Method for identifying passable areas in mines based on spatiotemporal continuous compensation

可通行区域识别是矿山无人驾驶技术中的重要环节。露天矿山道路场景具有道路边界模糊不清及路面平坦度不一等特征,使用传统同心圆地面分割模型进行矿山道路平面拟合时容易出现可通行区域与车辆不连通及帧间可通行区域识别结果不一致等误分类问题。提出了一种基于时空连续补偿的矿山道路可通行区域识别方法。首先,基于同心圆模型对矿山道路建模,并利用主成分分析方法进行多平面拟合,获取初始可通行区域分割结果;然后,基于空间连通性,分别利用区域生长方法和基于密度的噪声应用空间聚类方法对初始可通行区域进行区域连通性滤波及点连通性滤波,得到符合空间连通性的可通行区域;最后,基于时间区域一致性对不同点云帧中可通行性不一致的不稳定区域进行滤除,先根据正态分布变换方法构建栅格地图,再利用时间稳定权重判断栅格稳定性,最终通过区域栅格投影实现不稳定区域的滤除。矿山场景中测试结果表明:该方法的准确率为93.44%,较现有主流方法提升2.27%;召回率为99.14%,较现有主流方法提升8.26%。该方法不仅在不连通区域中具有良好的空间连通性,还在崎岖区域内具有良好的时序稳定性。

Identifying passable areas is a crucial aspect of autonomous driving technology in mining.Openpit mining road scenes are characterized by unclear road boundaries and varying surface flatness.When using traditional concentric circle ground segmentation models for fitting mining road planes,misclassification issues often arise,such as disconnection between passable areas and vehicles,and inconsistencies in passable area recognition results across frames.This paper proposed a method for identifying passable areas in mining roads based on spatiotemporal continuous compensation.First,the mining road was modeled using a concentric circle model,and principal component analysis was applied for multi-plane fitting to obtain the initial segmentation results of passable areas.Next,based on spatial connectivity,regional connectivity filtering and point connectivity filtering were performed on the initial passable areas using the region-growing algorithm and density-based spatial clustering of applications with noise algorithm,respectively,to obtain passable areas that meet spatial connectivity criteria.Finally,to eliminate unstable regions with inconsistent passability across different point cloud frames,a grid map was constructed based on a normal distribution transformation algorithm,and temporal stability weights were used to assess grid stability,ultimately filtering out unstable regions through regional grid projection.Test results in mining indicated that the proposed method for identifying passable areas achieved an accuracy of 93.44%,representing a 2.27%improvement over existing mainstream algorithms;the recall rate was 99.14%,reflecting an 8.26%enhancement compared to current mainstream algorithms.The proposed method not only exhibits good spatial connectivity in disconnected areas but also demonstrates strong temporal stability in rugged regions.