Results of the ISPRS benchmark on urban object detection and 3D building reconstructionby Franz Rottensteiner, Gunho Sohn, Markus Gerke, Jan Dirk Wegner, Uwe Breitkopf, Jaewook Jung

ISPRS Journal of Photogrammetry and Remote Sensing

Similar

Salient Object Detection: A Benchmark

Authors:
Ali Borji, Dicky N. Sihite, Laurent Itti
2012

Laparoscopic gastrectomies for cancer: The ACOI-IHTSC national guidelines

Authors:
Umberto Bracale, Giusto Pignata, Marco Maria Lirici, Cristiano GS Hüscher, Raffaele Pugliese, Giovanni Sgroi, Giovanni Romano, Giuseppe Spinoglio, Monica Gualtierotti, Valeria Maglione, Santiago Azagra, Eiji Kanehira, Jun GI Kim, Kyo Young Song, On behalf of the Guideline Committe and Italian Hi-Tech Surgical Club -
2012

Association Between Nontraditional Risk Factors and Metabolic Syndrome in Indigenous Argentinean Schoolchildren

Authors:
Valeria Hirschler, Gustavo Maccallini, Claudia Molinari, Inés M. Urrutia, Luis A. Castano, on behalf of the San Anton
2014

An Account of the State Prison, or Penitentiary House in New-York (Concluded)

Authors:
One of the Inspectors of the Prison
1811

Text

oc, urge 00 K

Net , Wo 3D building reconstruction

Aerial imagery es, by

Researchers were encouraged to submit their results of urban object detection and 3D building reconn obje porta decade puter vision, the success of the Middlebury Stereo Vision test (Scharstein and Szeliski, 2002) and other benchmarks such as the

Pascal VOC data set (Everingham et al., 2010) has shown the importance of providing common data sets with ground truth for comparing different approaches to problems such as image matching employ priors for at the sky is usuare most li ), whereas other hand, they cannot exploit features such as the NDVI th be extracted from images taken by modern multispectral s

There is an obvious need for benchmark data sets consisting borne data that can serve as test beds for developments in the field of topographic object detection and 3D reconstruction, in particular in urban areas.

There have been attempts in the past to distribute benchmark data sets for object extraction. The authors particularly acknowledge the efforts of OEEPE/EuroSDR (European Spatial Data

Research), who provided data sets for building (Kaartinen et al., 2005) and road extraction (Mayer et al., 2006) and for automated ⇑ Corresponding author.

E-mail addresses: rottensteiner@ipi.uni-hannover.de (F. Rottensteiner), gsohn@ yorku.ca (G. Sohn), m.gerke@utwente.nl (M. Gerke), jan.wegner@geod.baug.ethz.ch (J.D. Wegner), breitkopf@ipi.uni-hannover.de (U. Breitkopf), jwjung@yorku.ca

ISPRS Journal of Photogrammetry and Remote Sensing xxx (2014) xxx–xxx

Contents lists availab

ISPRS Journal of Photogramm els(J. Jung).As a consequence, the authors usually evaluate their methods on different data sets and using different evaluation criteria, which makes a comparison of the methods difficult and hampers a critical assessment of the pros and cons of each of the methods. In comthat are tailored to such data and, for instance, the location of objects in an image, modelling th ally the highest object in a scene and that roads the bottom of an image (Yang and Förstner, 20110924-2716/$ - see front matter  2013 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS). Published by Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.isprsjprs.2013.10.004

Please cite this article in press as: Rottensteiner, F., et al. Results of the ISPRS benchmark on urban object detection and 3D building reconstruction.

Photogram. Remote Sensing (2014), http://dx.doi.org/10.1016/j.isprsjprs.2013.10.004kely at on the at can ensors. of air-object extraction is still an active field of research, with the focus shifting to detailed representations of objects, to using data from new sensors, or to advanced processing techniques. However, scanning the relevant literature in photogrammetry and remote sensing (Schindler et al., 2011; Sohn et al., 2013; Stilla et al., 2011), it has become obvious that there is a lack of publicly available benchmark data sets with ground truth that can be used for the evaluation of their methods by the authors of research papers.

However, using standard benchmarks for object extraction from computer vision such as the Pascal VOC data set for a comparison of object extraction techniques from remote sensing imagery is not necessarily fair to the latter. Methods tailored for remote sensing data, usually characterised by vertical viewing directions, cannot rely on the availability of a reference direction such as the vertical in terrestrial images with horizontal viewing directions. Thus, on the one hand, they may perform poorly in comparison to methodsLaser scanning

Evaluation

Benchmarking test 1. Introduction

The automated extraction of urba by airborne sensors has been an im photogrammetry for more than twostruction, which were evaluated based on reference data. This paper presents the outcomes of the evaluation for building detection, tree detection, and 3D building reconstruction. The results achieved by different methods are compared and analysed to identify promising strategies for automatic urban object extraction from current airborne sensor data, but also common problems of state-of-the-art methods.  2013 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS). Published by Elsevier

B.V. All rights reserved. cts from data acquired nt topic of research in s (Mayer, 2008). Urban and object detection. Apart from making different approaches comparable, benchmarks can trigger progress by giving indications about the most promising strategies for the solution of a given task and by identifying common problems of existing approaches, thus showing new directions of research.Keywords:

Automatic object extraction comparable, benchmarking data sets are of paramount importance. Such a data set, consisting of airborne image and laserscanner data, has been made available to the scientific community by ISPRS WGIII/4.Results of the ISPRS benchmark on urban and 3D building reconstruction

Franz Rottensteiner a,⇑, Gunho Sohn b, Markus Gerke a Institute of Photogrammetry and GeoInformation, Leibniz Universität Hannover, Nienb bGeoICT Lab, Earth and Space Science and Engineering Department, York University, 47 c Faculty ITC, EOS Department, University of Twente, PO Box 217, 7500AE Enschede, The d Institute of Geodesy and Photogrammetry, Swiss Federal Institute of Technology Zurich a r t i c l e i n f o

Article history:

Available online xxxx a b s t r a c t

For more than two decad objects from data acquired journal homepage: www.bject detection

Jan Dirk Wegner d, Uwe Breitkopf a, Jaewook Jung b r Straße 1, 30167 Hannover, Germany eele St., Toronto M3J 1P3, Canada herlands lfgang-Pauli-Strasse 15, 8093 Zurich, Switzerland many efforts have been made to develop methods for extracting urban airborne sensors. In order to make the results of such algorithms more le at ScienceDirect etry and Remote Sensing evier .com/ locate/ isprs jprsISPRS J. on urban object extraction. A modern data set consisting of digital aerial image and ALS data along with reference data was generated to acquire the ALS data at a flying height of 650 m in 6 strips with a 2 ammand made available to the research community via the ISPRS web site (ISPRS, 2013). Unlike previous benchmark data sets on urban object detection, the reference data include 2D outlines of multiple object types and 3D roof landscapes. It also contains different types of urban development. Researchers are given access to the sensor data and encouraged to carry out one or more of several urban object extraction tasks: