Home | english  | Impressum | KIT

Software and Datasets

3D Scene Analysis

 

Neighborhood Selection (Source optNESS.zip)

This is our tool optNESS (optimal NEighborhood Size Selection) for deriving optimal 3D neighborhoods via eigenentropy-based scale selection. Detailed instructions on how to use the tool are provided in readme.txt. If you make use of our tool, please refer to the corresponding publication:

M. Weinmann, B. Jutzi, and C. Mallet (2014): Semantic 3D scene interpretation: a framework combining optimal neighborhood size selection with relevant features. In: K. Schindler and N. Paparoditis (Eds.), ISPRS Technical Commission III Symposium. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. II-3, pp. 181-188. PCV 2014 Best Paper Honorable Mention.

Feature Extraction (Source geoFEX.zip)

This is our tool geoFEX (geometric Feature EXtraction) for extracting low-level geometric 2D and 3D features. Detailed instructions on how to use the tool are provided in readme.txt. If you make use of our tool, please refer to the corresponding publication:

M. Weinmann, S. Urban, S. Hinz, B. Jutzi, and C. Mallet (2015): Distinctive 2D and 3D features for automated large-scale scene analysis in urban areas. Computers & Graphics, Vol. 49, pp. 47-57.

Licence: GNU GPLv2
Contact: Dr.-Ing. M. Weinmann
top

 

Fisheye calibration image data set

This data set contains fisheye images of a checkerboard recorded with different lenses and sensors.  
Licence: GNU GPLv2
Contact: Dipl.-Ing. S. Urban
Source: https://github.com/urbste/ImprovedOcamCalib
top

 

The EnMAP Contest

The dataset for the EnMAP Contest originates from the simulated EnMAP Alpine Foreland image covering the area around the Ammersee in Bavaria, Germany. Based on the original dataset, 20 different land use classes have been defined on the screen by focussing on visual differences in the images (considering several channel combinations) but also by checking pixels’ individual spectra.
Licence: GNU GPLv2
Publication: Braun, A.C.; Weinmann, M.; Keller, S.; Müller, R.; Reinartz, P.; Hinz, S. (2015): The EnMAP contest: developing and comparing classification approaches for the Environmental Mapping and Analysis Programme - Dataset and first results. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. XL-3/W3, pp. 169-175. [pdf]
Contact: Dipl.-Ing. M. Weinmann
Source: EnMAP_Contest.zip (1.57 GB)
Instructions for participating in the EnMAP Contest

(pdf, 26 kB)

Exemplary MATLAB-script for classification (zip, 1 kB)

Evaluation: The evaluation procedure will be similar to the ones of the Middlebury Stereo Vision benchmark, the ISPRS Test Project on Urban Classification, 3D Building Reconstruction and Semantic Labeling and the IQmulus & TerraMobilita Contest, i.e. if you are interested in participating, please submit the achieved quality measures and the classified image. After submitting these data, your results will be evaluated and you can request that your results are added to the permanent table. For the latter, it is mandatory to provide (1) a short description of the algorithm or the reference to a paper where this algorithm has been published and (2) respective code fragments or hints on the involved software packages.
Deadline of the first EnMAP Contest: 15th Feburary 2016
top

Improved Wide-Angle, Fisheye and Omnidirectional Camera Calibration

This is an add-on to the online available omnidirectional camera calibration toolbox OCamCalib of Scaramuzza et al. The 3 image data sets necessary to run all test calibrations can be found below. Please consult the readme.txt on how to install the add-on. If you make use of our extended version, please cite the corresponding paper.  
Licence: GNU GPLv2
Publication: Urban, S.; Leitloff, J.; Hinz, S. (2015): Improved Wide-Angle, Fisheye and Omnidirectional Camera Calibration. ISPRS Journal of Photogrammetry and Remote Sensing 108, 72-79.
Contact: Dipl.-Ing. S. Urban
Source: https://github.com/urbste/ImprovedOcamCalib
top

 

Tracking Vehicles in Aerial Image Sequences

The following data set comprises aerial image sequences and a xml files with manually labeled trajectories of all visible vehicles.
The data set on this page is available under the following license: Creative Commons Lizenzvertrag

If you use the data, please refer to this web page or to the KIT AIS Data Set.

 

Many thanks to the German Aerospace Center  for providing the aerial image sequences.

DLR-Logo

 
Licence:  Creative Commons Lizenzvertrag
Publication:
Contact: Dipl.-Ing. F. Schmidt
Source: Training sequences  --- C++ code  (for easy import and further processing of the AIS CVML files)
top

 

Tracking People in Aerial Image Sequences

The following datasets have been created for the research project VABENE - People Traing in Aerial Image Sequences. Each dataset comprises of an aerial image sequence and a xml file with manually labeled trajectories of all visible persons.

All datasets on this page are available under the following license: Creative Commons Lizenzvertrag
If you use the data, please refer to this web page or to the KIT AIS Data Set.

 

 
Licence:  Creative Commons Lizenzvertrag
Publication:
Contact: Dipl.-Ing. F. Schmidt
Source: click here for further information
top

 

Workflow of LaFiDa
Workflow of LaFiDa
rigidbody labeling
Helmet with rigid body labels

LaFiDa - A Laserscanner Multi-Fisheye Camera Dataset

We provide test data for applying Simultaneous Localization and Mapping (SLAM) algorithms. The Laserscanner Multi-Fisheye Camera Dataset (LaFiDa) includes fully synchronized laserscanner data and image sequences for three diverge fisheye cameras, aswell as ground truth trajectory.

Each dataset consist of:

  • three image sequences, each sequence captured with a fisheye camera
  • the laserscanner data
  • an interactive 3D point cloud of the laserscanner data
  • the data conventions
  • a video of the scene to get first impressions
  • a link to download the complete test data

LaFiDa is supported by master students of Geodesy and Geoinformatics at KIT during a student project in "Project Computer Vision".

Note: LaFida is currently in a release candidate state. The use of LaFiDa is for free, but please refer to the publication below.

Licence: Creative Commons Lizenzvertrag

Contact: 

PD Dr.-Ing. Jutzi, Boris
Dr.-Ing. Urban, Steffen

Source: click here for further information.

Publication: S. Urban, B. Jutzi (2017) LaFiDa - A Laserscanner Multi-Fisheye Camera Dataset. Special Issue "3D Imaging". Open Access Journal of Imaging 2017, 3(1), 5 [doi: 10.3390/jimaging3010005]

top

 


MLPnP – A maximum likelihood solution to the Perspective-N-Point problem

MLPnP offers the possibility to integrate observation uncertainty into the direct solution of the PnP problem. This is especially useful if an estimate of the observation covariance is available. Given that estimate MLPnP outperforms the given state-of-the-art PnP solutions (see Figure on the right). If equal observation uncertainty is assumed MLPnP still outperforms the best polynomial solvers in terms of speed and accuracy.

The C++ version of MLPnP is integrated in a fork of OpenGV. You can find the link to the fork below. In addition we provide Matlab Code which is slower, but produces the same results. The toolbox provides all PnP solvers in a Matlab toolbox to reproduce the results presented in the paper.
 
Licence: 
Publication: Urban, S.; Leitloff, J.; Hinz, S. (2016): MLPnP - A Real-Time Maximum Likelihood Solution to the Perspective-n-Point Problem. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences. (pdf)
Contact: Dipl.-Ing. S. Urban
Source C++
Source Matlab
Source Matlab Toolbox
top

 

MultiCol-SLAM:  A Multi-Fisheye Camera System SLAM System

MultiCol-SLAM is a full multi-fisheye camera SLAM system build upon ORB-SLAM. It extends ORB-SLAM with some novel concepts and methods. First, keyframes are exchanged by multi-keyframes modelling the fact, that multiple images exist at each keyframe position. The multi-camera system is modeled using MultiCol and the camera projections within these model are calibrated using a generic polynomial model. Bundle adjustment is formulated as an optimization of a hyper graph and implemented in g2o. Furthermore, new types of descriptors are included, i.e. a distorted and an online learned, masked version of BRIEF. The loop closing mechanisms and re-localization capabilites of ORB-SLAM are extended for multi-camera systems.
 
Licence: GNU GPLv3
Publication: Urban, S.; Wursthorn, S.; Leitloff, J.; Hinz, S. (2016): MultiCol Bundle Adjustment: A Generic Method for Pose Estimation, Simultaneous Self-Calibration and Reconstruction for Arbitrary Multi-Camera Systems./ International Journal of Computer Vision (IJCV), pages 1-19, 2016.
Contact: Dipl.-Ing. S. Urban
Source C++
top