Lidar‐based city objects detection is an interesting topic along with the development of Laser scan equipment which has been widely applied in various applications such as 3D building reconstruction, navigation, and so on. In this article, we describe a city object detection algorithm for airborne Lidar images using superpixel segmentation and DenseNet classification. Compared with the existing studies, this article has two innovations. First, a DenseNet‐based city object classification model is trained by data sets automatically labeled from the OpenStreetMap. Second, the graph analysis is applied to further improve the classification of the superpixels. The results from an experiment in the London area indicate that the DenseNet‐based classification model trained by OpenStreetMap data can achieve 86% classification accuracy for building objects. With the proposed graph analysis, the detection accuracy of building objects increased to 98.5% in the test areas. Also, we testified that by dividing the city area into different types such as commercial, residential, and rural, the detection accuracy can be further improved. Based on the extensive examinations, it is suggested that the proposed superpixel classification method can be used to detect city objects from large‐scale low‐resolution Lidar image data (50 cm).

中文翻译：

---

使用 OpenStreetMap 标记的超像素从机载激光雷达数据中检测城市目标

随着激光扫描设备的发展，基于激光雷达的城市物体检测是一个有趣的话题，激光扫描设备已广泛应用于各种应用，如 3D 建筑重建、导航等。在本文中，我们描述了一种使用超像素分割和 DenseNet 分类的机载激光雷达图像城市目标检测算法。与已有研究相比，本文有两点创新。首先，基于 DenseNet 的城市对象分类模型通过从 OpenStreetMap 自动标记的数据集进行训练。其次，应用图分析进一步改进超像素的分类。伦敦地区的一项实验结果表明，由 OpenStreetMap 数据训练的基于 DenseNet 的分类模型可以实现 86% 的建筑对象分类准确率。通过提出的图形分析，在测试区域中建筑物物体的检测精度提高到 98.5%。此外，我们证明通过将城市区域划分为商业、住宅和农村等不同类型，可以进一步提高检测精度。基于广泛的研究，建议所提出的超像素分类方法可用于从大规模低分辨率激光雷达图像数据（50 cm）中检测城市对象。