Deep Neural Networks for Determining the Parameters of Buildings from Single-Shot Satellite Imagery
- PDF / 1,431,990 Bytes
- 13 Pages / 612 x 792 pts (letter) Page_size
- 80 Downloads / 209 Views
FICIAL INTELLIGENCE
Deep Neural Networks for Determining the Parameters of Buildings from Single-Shot Satellite Imagery A. N. Trekinb,c,**, V. Yu. Ignatieva,b,*, and P. Ya. Yakubovskiib,*** a Institute
of Control Sciences, Federal Research Center “Computer Science and Control,” Russian Academy of Sciences, Moscow, Russia b
Skolkovo Institute of Science and Technology, Moscow, Russia c
AEROCOSMOS Research Institute, Moscow, Russia *e-mail: [email protected] **e-mail: [email protected] ***e-mail: [email protected]
Received April 6, 2020; revised April 30, 2020; accepted May 25, 2020
Abstract—The height of a building is a basic characteristic needed for analytical services. It can be used to evaluate the population and functional zoning of a region. The analysis of the height structure of urban territories can be useful for understanding the population dynamics. In this paper, a novel method for determining a building’s height from a single-shot Earth remote sensing oblique image is proposed. The height is evaluated by a simulation algorithm that uses the masks of shadows and the visible parts of the walls. The image is segmented using convolutional neural networks that makes it possible to extract the masks of roofs, shadows, and building walls. The segmentation models are integrated into a completely automatic system for mapping buildings and evaluating their heights. The test dataset containing a labeled set of various buildings is described. The proposed method is tested on this dataset, and it demonstrates the mean absolute error of less than 4 meters.
DOI: 10.1134/S106423072005007X
INTRODUCTION Determining the geometric parameters of buildings from satellite images is an important task performed by map services in the process of manual image interpretation. This process requires highly skilled experts and significant time for interpreting images of large areas. Some parameters, such as the object contours, their area and height can be determined very accurately, which offers possibilities of applying modern neural network approaches to develop object recognition algorithms and automate the procedure of reconstructing 3D models of objects, such as buildings. A building’s height is an important characteristic that must be taken into account in mapping and analyzing urban areas. Given the height and area of a residential building, the number of its residents can be evaluated. The map of the building’s height can also be used for the semiautomatic evaluation of the building’s footprints from oblique images, monitoring construction progress, and other purposes. The first automated algorithms for determining the height of an object from remote sensing data used either stereo pair images or a single image [1]. Presently, this task can be performed completely automatically. An example of the algorithm for evaluating the building height from remote sensing data based on photogrammetry can be found in [2]. It demonstrates high accuracy but requires costly data—stereo pairs of satell
Data Loading...
