Building Segmentation and Classification from Aerial LiDAR via Local Planar Features
In this paper, we propose a framework on building segmentation and classification from Aerial Lidar data via planar features. In this framework, the planar points corresponding to planar objects are obtained first by an unsupervised Markov random field cl
- PDF / 2,685,527 Bytes
- 10 Pages / 439.37 x 666.142 pts Page_size
- 82 Downloads / 194 Views
Institute of Computer Vision, Nanchang Hangkong University, Nanchang, China [email protected] 2 Key Laborator of Jiangxi Province for Image Processing and Pattern Recongnition, Nanchang, China 3 NLPR, Institute of Automation, Chinese Academy of Sciences, Beijing, China 4 National Disaster Reduction Center of China, Ministry of Civil Affairs, Beijing, China
Abstract. In this paper, we propose a framework on building segmentation and classification from Aerial Lidar data via planar features. In this framework, the planar points corresponding to planar objects are obtained first by an unsupervised Markov random field clustering model. The ground normal is detected from planar points via the proposed constrained K-means algorithm. Within constrained K-means algorithm, the building points are generated by removing ground points from planar points. Furthermore, the candidate buildings are obtained by using region growing algorithm. Finally, these candidate buildings are classified into two types, that is, abnormal building and normal building based on the proposed vertical feature. Experimental results on a real world dataset demonstrate the effectiveness of our framework. Keywords: Building segmentation · Planar objects · Aerial lidar data · Ground detection
1
Introduction
Earthquake and flood have taken place frequently over the world and brought disasters to natives. Many buildings are collapsed and damaged in the affected areas. In practice, it is difficult to measure and evaluate the damaged condition of buildings by manpower. Many algorithms based on computer vision have been proposed, among which image-based approaches are widely used. However, image-based approaches may not be applied to real-world problems effectively because the image acquisition is susceptible to lighting conditions. By contrast, 3D point cloud is robust to light. Hence, we propose a method to detect and classify buildings from 3D point cloud. J. Chu—This work was supported in part by the National Natural Science Foundation of China (Grant Nos. 61263046, 61403376, 61175025 and 41301485). c Springer-Verlag Berlin Heidelberg 2015 H. Zha et al. (Eds.): CCCV 2015, Part II, CCIS 547, pp. 313–322, 2015. DOI: 10.1007/978-3-662-48570-5 31
314
J. Chu et al.
Fig. 1. A framework of building detection and classification.
1.1
Related Work
Building detection and classification approaches can be mainly divided into two categories, i.e., supervised learning based methods [1–4], and unsupervised learning based methods [5–9]. For the supervised learning based approaches, the features for points classification are learned by fitting a mixture of Gaussian model by Charaniya et al. [1] and Lalonde et al. [2] [3]. Secord et al. [4] proposed a method based on support vector machines for object detection using aerial lidar and image data.
Building Segmentation and Classification from Aerial LiDAR
315
Fig. 2. One segmentation and classification result on data1. In sub-figures (b) and (c), each color represents a normal building except the red color. The red color repres
Data Loading...
