利用YOLO深度学习算法识别复杂道路交叉口  

Recognition of complex road junctions using YOLO deep learning algorithm

作　　者：马超 刘平芝[1,2] 熊顺 徐道柱[1,2] MA Chao;LIU Pingzlii;XIONG Shun;XU Daozhu(Xi'an Research Institute of Surveying and Mapping,Xi'an 710054;China 2.State Key Laboratory of Geo-Information Engineering,Xi’an 710054,China)

机构地区：[1]西安测绘研究所,陕西西安710054 [2]地理信息工程国家重点实验室,陕西西安710054

出　　处：《测绘科学与工程》2019年第4期24-29,共6页Geomatics Science and Engineering

基　　金：国家自然科学基金(多源矢量空间数据集成与融合理论和方法研究,基金编号:41571399);地理信息工程国家重点实验室自主研究课题(编号:SKLGIE2018-ZZ-7)。

摘　　要：针对目前道路复杂交叉口基于几何形状描述与图形匹配识别方法存在的不足,引入了计算机视觉领域中的目标检测方法,利用YOLO深度学习算法识别复杂道路交叉口。首先,将矢量数据栅格化,并通过人工标注的方法得到样本集,为了提高样本质量,对样本的比例尺和符号化进行设计;然后,利用该样本集对YOLO神经网络进行训练,训练后的网络即可应用于复杂道路交叉口的识别。试验表明,该方法可以有效提高复杂道路交叉口识别的敷率。Aiming at the deficiency of the existing road junction recognition methods based on geometric shape description and shape matching,an object detection rnelluxl in computer vision is introduced and the YOLO deep learning algorithm is used for recognition.Firstly,the vector data has been rasterized,and the sample set is obtained after manually annotated.The scale and symbolic representation methods are designed to improve the sample quality.Secondly,the YOLO neural network has been trained with the sample set ainl used to recognize the complex road junctions.The experinienlal results show that this nirlliod can effectively increase the recognition efficiencv.

关 键 词：道路结构识别 道路交叉口 YOLO算法 目标识别 计算机视觉 

分 类 号：P285[天文地球—地图制图学与地理信息工程]