Omnidirectional images database

INTRODUCTION

THE DATABASE

PUBLICATIONS

VISIT ALSO


Introduction

Quorum V is a set of omnidirectional images captured in an indoors environment (Quorum V building, 2nd floor) at Miguel Hernandez University, Spain. This database is intended to test visual mapping and localization algoritmhs for mobile robots.

The database has been referenced in the following papers:

  1. Performance of Global-Appearance Descriptors in Map Building and Localization Using Omnidirectional Vision
    L. Payá, F. Amorós, L. Fernández, O. Reinoso
    Sensors. Vol. 14, pp. 3033-3064. (February 2014).
    Ed. MDPI. ISSN:1424-8220. DOI: 10.3390/s140203033

The database includes a corridor, three offices, a library and an events room. It is composed of two sets of images.

  • Set 1: Training set. It is composed of 872 omnidirectional color images which have been captured on a dense regular 40x40 cm. grid of points.
  • Set 2: Test set. It is composed of 546 images captured in all the rooms, in some half-way points among the grid points and with different orientations and times of day.

The next figure shows a bird eye's view of the grid points where the robot captured the training set.

Omnidirectional images database

Room 1: Corridor. 212 images.

Room 2: Office A. 35 images.

Room 3: Office B. 72 images.

Room 4: Office C. 84 images.

Room 5: Library. 169 images.

Room 6: Events room. 300 images.

 

 

The database

The whole database is downloadable through the next links:

Quorum V database
Training set Omnidirectional images Panoramic images
Test set Omnidirectional images Panoramic images

The database was captured with an Imaging Source DFK 21BF04 camera, which takes pictures of a hyperbolic mirror (Eizoh Wide 70). The mirror is mounted over the camera, with its axis aligned with the camera optic axis. Both the original omnidirectional images, and the panormic images (cylindrical projection) can be downloaded.

The next figures shows a bird eye's view of each room. The coordinates where each image was captured are shown. Red squares are the positions where training images were captured and green crosses the positions of test images. On each green cross, 16 test images were captured, with different orientations (every 22.5 deg). The name of the training images contains the coordinates of the capture points (cm) and the name of the test images contains the coordinates and the orientation of the camera (deg).

 

Room 1: Corridor

 

 

Room 6: Events room

Omnidirectional images. Room 1: Corridor   Omnidirectional images. Room 5: Events' room

 

Room 2: Office A

 

 

Room 3: Office B

Omnidirectional images. Room 2: Office A   Omnidirectional images. Room 3: Office B

 

Room 4: Office C

 

 

Room 5: Library

Omnidirectional images. Room 4: Office C   Omnidirectional images. Room 5: Library

Publications

  1. Performance of Global-Appearance Descriptors in Map Building and Localization Using Omnidirectional Vision
    L. Payá, F. Amorós, L. Fernández, O. Reinoso
    Sensors. Vol. 14, pp. 3033-3064. (February 2014).
    Ed. MDPI. ISSN:1424-8220. DOI: 10.3390/s140203033

Visit also

  1. ARVC (Automation, Robotics and Computer Vision group): Visit our research group.