面向无人驾驶的野外道路可通行区域检测  

Detection of Accessible Area of Field Road for Ummanned Driving

作　　者：陈湘 武富春[1] 王明[1] 张耀 郝宇欣 潘立雪 CHEN Xiang;WU Fuchun;WANG Ming;ZHANG Yao;HAO Yuxin;PAN Lixue(North Automatic Control Technology Institute,Taiyuan 030006,China)

机构地区：[1]北方自动控制技术研究所,太原030006

出　　处：《火力与指挥控制》2023年第4期84-89,共6页Fire Control & Command Control

摘　　要：针对野外非结构化道路中存在的环境背景复杂、道路类型多变、没有清晰车道线、没有规则道路边界等难点,提出一种基于语义分割的可通行区域检测方法。对所选取的数据集进行处理,将数据集标注为可通行区域、难通行区域、不可通行区域、天空4种类型;应用双边语义分割网络Bisenet进行4种类型的分割;再应用C-均值聚类算法进行可通行区域部分的提取以及二次分割,进而实现更精细的可通行区域检测。通过验证,该方法具有良好的检测效果,平均交并比达到78%,高于传统的方法。在野外道路无人驾驶方面具有一定的实际应用价值。As the environment background is complex,the types of roads are variable,there are no clear lane markings and no regular road boundaries and there are other difficulties existing in the unstructured roads in the field,a method for detecting the accessible area based on semantic segmentation is proposed.Firstly,the selected data set is processed,and the data set is labeled as accessible area,difficultly accessible area,unaccessible area and sky;secondly,the bilateral semantic segmentation network Bisenet is used for the segmentation of four types,then,C-means clustering algorithm is applied to extract the part of the accessible area and to perform the secondary segmentation,so as to achieve more precise detection of the accessible area.Through verification,the proposed method has a good detection effect,and the Mean Intersection over Union reaches 78%,which is higher than that of the traditional method.It has certain practical application value for the unmanned driving in the field road.

关 键 词：聚类 深度学习 无人驾驶 语义分割 

分 类 号：TJ811[兵器科学与技术—武器系统与运用工程]