Robotics Program, University of Michigan, Ann Arbor, MI 48109, USA.
Sensors (Basel). 2018 Nov 15;18(11):3960. doi: 10.3390/s18113960.
Geographic information systems (GIS) provide accurate maps of terrain, roads, waterways, and building footprints and heights. Aircraft, particularly small unmanned aircraft systems (UAS), can exploit this and additional information such as to improve navigation accuracy and safely perform contingency landings particularly in urban regions. However, building roof structure is not fully provided in maps. This paper proposes a method to automatically label building roof shape from publicly available GIS data. Satellite imagery and airborne LiDAR data are processed and manually labeled to create a diverse annotated roof image dataset for small to large urban cities. Multiple convolutional neural network (CNN) architectures are trained and tested, with the best performing networks providing a condensed feature set for support vector machine and decision tree classifiers. Satellite image and LiDAR data fusion is shown to provide greater classification accuracy than using either data type alone. Model confidence thresholds are adjusted leading to significant increases in models precision. Networks trained from roof data in Witten, Germany and Manhattan (New York City) are evaluated on independent data from these cities and Ann Arbor, Michigan.
地理信息系统（GIS）提供精确的地形、道路、水路和建筑物轮廓及高度地图。飞机，特别是小型无人机系统（UAS），可以利用这些信息以及其他信息，如 ，提高导航精度，并安全地进行应急着陆，特别是在城市地区。然而，建筑物的屋顶结构在地图中并没有完全提供。本文提出了一种从公开的 GIS 数据中自动标记建筑物屋顶形状的方法。对卫星图像和机载激光雷达数据进行处理和手动标记，为小到大城市创建了一个多样化的注释屋顶图像数据集。训练和测试了多个卷积神经网络（CNN）架构，表现最好的网络为支持向量机和决策树分类器提供了一个精简的特征集。与单独使用任何一种数据类型相比，卫星图像和激光雷达数据融合显示出更高的分类准确性。调整模型置信度阈值可显著提高模型的精度。在德国维滕和曼哈顿（纽约市）的屋顶数据上训练的网络在这些城市和密歇根州安阿伯的独立数据上进行了评估。
