Abstract | ||
---|---|---|
Identifying housing buildings from afar is required for many urban planning and management tasks, including population estimations, risk assessment, transportation route design, market area delineation and many decision making processes. High-resolution remote sensing provides a cost-effective method for characterizing buildings and, ultimately, determining its most likely use. In this study we combined high-resolution multispectral images and LiDAR point clouds to compute building characteristics at the parcel level. Tax parcels were then classified in one of four classes (three residential classes and one non-residential class) using three classification methods: Maximum likelihood classification (MLC), Suport Vector Machines (SVM) with linear kernel and SVM with non-linear kernel. The accuracy assessment from a random sample showed that the maximum MLC was the most accurate method followed by SVM with linear kernel. The best classification method was then applied to the whole study area and the residential class was used to mask-out non-residential buildings from a building footprint layer. |
Year | DOI | Venue |
---|---|---|
2014 | 10.1007/978-3-319-07491-7_39 | PATTERN RECOGNITION, MCPR 2014 |
Keywords | Field | DocType |
Remote sensing, LiDAR, housing units, land use classification | Kernel (linear algebra),Population,Computer science,Support vector machine,Multispectral image,Remote sensing,Lidar,Sampling (statistics),Artificial intelligence,Footprint,Point cloud,Machine learning | Conference |
Volume | ISSN | Citations |
8495 | 0302-9743 | 0 |
PageRank | References | Authors |
0.34 | 5 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
José Luis Silván-Cárdenas | 1 | 10 | 3.34 |
Juan Andrés Almazán-González | 2 | 0 | 0.34 |
Stephane Couturier | 3 | 5 | 1.89 |