摘要
提出了一种基于单目视觉和激光雷达的同步定位与制图融合语义信息的新方法,该方法利用从单目视觉中提取的特征和在激光雷达深度地图中的对应关系来获得相对于关键帧的姿态数据,同时对图像进行语义分割;通过从深度卷积神经网络CNN获得的语义特征来细化语义信息,地图中的每个点都与一个语义特征相关联,以执行语义引导的本地和全局姿态优化。提出的语义标记和SLAM的耦合具有更好的鲁棒性和准确性。在室内环境中对装备单目视觉和激光雷达的移动机器人进行验证实验,实验结果表明:该方法可以提高机器人导航精度,实现机器人智能自主导航,同时也可以提供语义信息的图像数据。
A synchronous positioning and mapping method combining semantic information based on monocular vision and lidar was proposed.The pose data relative to the keyframe were obtained based on the features extracted from the monocular vision and the correspondence in the lidar depth map,and the image was semantically segmented.Semantic information was refined by the semantic features obtained from deep convolutional neural network(CNN),and each point in the map was associated with a semantic feature to perform local and global pose optimization by semantically guiding.The coupling of proposed semantic markup and SLAM is robust and accurate.Mobile robots equipped with monocular vision and lidar were tested in indoor environments.The experimental results show that the proposed method can improve the navigation accuracy of robot,realize intelligent autonomous navigation of robot,and also provide image data with semantic information.
作者
陈国军
陈巍
郁汉琪
王涵立
CHEN Guojun;CHEN Wei;YU Hanqi;WANG Hanli(Industrial Center,Nanjing Institute of Technology,Nanjing Jiangsu 211167,China)
出处
《机床与液压》
北大核心
2020年第9期16-20,共5页
Machine Tool & Hydraulics
基金
教育部产学合作协同育人项目(201801166003,201802082010)
南京市产学研合作资助项目(221722072)
南京工程学院产学研前瞻性项目(CXY201916)
南京工程学院挑战杯培育项目(TP20190004)。
关键词
SLAM
语义分割
CNN
SLAM
Semantic segmentation
Convolutional neural network(CNN)
