The Málaga Stereo and Laser Urban Data Set

The Málaga Urban Dataset was gathered entirely in urban scenarios with a car equipped with one stereo camera (Bumblebee 2) and five laser scanners, among other sensors. Its distinctive feature is the existence of high-resolution stereo images grabbed at a high rate (20 fps) during a 36.8 km trajectory, which turns it into a suitable benchmark for a variety of computer vision techniques, such as visual odometry, visual SLAM, loop closure and place recognition. Both plain text and binary files are provided, as well as open source tools for working with the binary versions.

Málaga Urban Dataset: stereo image from the car, GPS path over a city map and laser point cloud

Citation and license

Please cite this paper if you use the dataset:

J.L. Blanco-Claraco, F.A. Moreno-Dueñas, J. Gonzalez-Jimenez. “The Málaga Urban Dataset: High-rate Stereo and Lidars in a realistic urban scenario”. The International Journal of Robotics Research (IJRR), 2014, vol. 33, no. 2, pp. 207-214. DOI: 10.1177/0278364913507326 (draft PDF)

BibTeX 
@article{blanco2014malaga,
  author    = {Blanco-Claraco, Jos{\'e}-Luis and Moreno-Due{\~n}as, Francisco-{\'A}ngel and Gonz{\'a}lez-Jim{\'e}nez, Javier},
  title     = {The M{\'a}laga Urban Dataset: High-rate Stereo and Lidars in a realistic urban scenario},
  journal   = {The International Journal of Robotics Research},
  volume    = {33},
  number    = {2},
  pages     = {207--214},
  year      = {2014},
  doi       = {10.1177/0278364913507326}
}

This dataset is copyright by us and published under the Creative Commons Attribution CC BY 4.0 License.

Downloads

  • Selected extracts: 15 short sections of the dataset, much easier to work with than the entire data. See below.
  • Entire dataset: plain text, binary and image packages. See further below.
  • Alternative location: the entire dataset is also available in Google Drive, in case the main server is unavailable.

Selected extracts

All packages include raw and rectified stereo images, and all data in plain text files and in binary (.rawlog) format. The format is described in the extracts readme. The path of each extract is drawn from its GPS data, with start (green) and end (black) points.

Entire dataset

PackageContentsSizeMD5
Plain text Raw laser scans, IMU and GPS data, plus a list of all image file names. See the format readme included. 1.5 GB 8806e8c6b7b032310d41cd231d956dca
Binary (.rawlog) The main binary log with all sensors (malaga-urban-dataset_all-sensors.rawlog), a filtered camera + GPS version, and a KML of the path. Open with RawLogViewer, part of MRPT. Raw images are not included in the rawlogs: download them below and place them in a directory named "Images". 1.4 GB 32cc0a4f49116ca4d585a7f4e7ddc1e7
Raw images, 1024x768, 20 fps Unprocessed stereo images in JPEG format (separate left and right files) and the calibration parameters. 47.7 GB 5dad0ce7c77aa56d7f86ea16c200d5f4
Rectified images, 1024x768, 20 fps Rectified stereo images, calibration parameters and a camera + GPS rawlog. 41.5 GB d96e84ccd5aab6a3446364b24aafa3ee
Rectified images, 800x600, 20 fps Rectified stereo images, calibration parameters and a camera + GPS rawlog. 28.7 GB 88ba785e9d317d01127c13ffd9a0e7c7

Calibration parameters of the cameras: raw images, rectified 1024x768 and rectified 800x600. Also available: the calibration and data sheet of the MTi IMU.

Other files

PackageContentsSizeMD5
Stereo calibration images Image pairs of a checkerboard grabbed with the Bumblebee2 camera, useful to determine the camera rectification parameters. 10 MB
Video: stereo left (AVI) Left camera video of the entire dataset. 3.5 GB 10b71137e0a08f785c14dcd27475e096
Video: stereo right (AVI) Right camera video of the entire dataset. 3.5 GB 40026fb261b00c64ec0fa29c726f9558
Video: dataset preview (FLV) Composite preview: camera, GPS position over a city map and a simple 3D reconstruction. 5.6 GB dcefbf5baf62dd6df3f9eccee9fc4f21

All .rawlog files can be opened with the program RawLogViewer, part of MRPT. In Ubuntu, install the package mrpt-apps. Raw (not undistorted) images are not included in the rawlogs: download them separately and put them in a directory named Images for the software to find them.

Sensors

The vehicle is a Citroën C4 with all the sensors installed in a modified roof rack, so it can drive among the city traffic without restrictions.

Sensor Description Rate
Stereo camera Point Grey Bumblebee 2, color, 1024x768 20 Hz
Laser scanners (3) Hokuyo UTM-30LX: front (tilted down), left and right (vertical) 40 Hz
Laser scanners (2) SICK LMS-200: front and rear (horizontal) about 75 Hz
IMU Xsens MTi 100 Hz
GPS One receiver, used for the paths shown here about 2 Hz

Overall, the dataset has 113,082 stereo frames over 5,655 seconds (about 1.5 hours) and 2.15 million sensor observations. The sensor labels in the logs are CAMERA1, GPS_DELUO, HOKUYO1, HOKUYO2, HOKUYO3, LASER1, LASER2 and XSensMTi.

The car used to record the Málaga Urban Dataset, after installing the sensors
The car after installing the sensors.
Sensors of the Málaga Urban Dataset before mounting them on the car
Sensors before mounting on the car.

Overview of the entire dataset

Path

The path obtained from the GPS sensor during the whole dataset run, about 36 km through different urban areas of Málaga (KML file for Google Earth).

GPS path of the entire Málaga Urban Dataset, about 36 km

Video

This composition shows: the raw images from the left stereo camera (top-left), the GPS position over a map of the city (bottom-left) and a 3D reconstruction of the environment by simple interpolation of GPS points, using two of the five lasers (right). The exact timestamp is displayed, which is useful to skip to an interesting part for your application.

The video can also be downloaded for offline usage: see Video: dataset preview (FLV) in the table above.

Parsing the logs from C++

Apart from plain text files, binary .rawlog files are an efficient and convenient way of parsing vision and robotic datasets from C++ code. The following example program is a starting point to write your own: parse-dataset-example. Example of usage:

./parse_dataset_example malaga-urban-dataset-extract-01/malaga-urban-dataset-extract-01_rectified_800x600.rawlog

Analysis of the rawlog file

Output of rawlog-edit --info -i malaga-urban-dataset_all-sensors.rawlog for the entire dataset:

Physical file size                : 1.41 GB
Uncompressed file size            : 4.79 GB
Overall number of objects         : 2153717
All sensor labels                 : CAMERA1, GPS_DELUO, HOKUYO1, HOKUYO2, HOKUYO3, LASER1, LASER2, XSensMTi
Sensor (Label/Occurs/Rate/Durat.) :         CAMERA1 / 113082 /19.998 /5654.574
Sensor (Label/Occurs/Rate/Durat.) :       GPS_DELUO /  11244 /1.989 /5653.000
Sensor (Label/Occurs/Rate/Durat.) :         HOKUYO1 / 225416 /39.864 /5654.617
Sensor (Label/Occurs/Rate/Durat.) :         HOKUYO2 / 225631 /39.902 /5654.624
Sensor (Label/Occurs/Rate/Durat.) :         HOKUYO3 / 225510 /39.880 /5654.621
Sensor (Label/Occurs/Rate/Durat.) :          LASER1 / 398531 /74.974 /5315.578
Sensor (Label/Occurs/Rate/Durat.) :          LASER2 / 404487 /73.568 /5498.109
Sensor (Label/Occurs/Rate/Durat.) :        XSensMTi / 549816 /100.000 /5498.150

Changelog

  • 2026-11-02: New home of the dataset page (previously at mrpt.org/MalagaUrbanDataset).
  • 2019-01-29: The downloads server had technical problems. An alternative location in Google Drive was added.
  • 2018-09-13: Fix typo in the "z" coordinate of the front stereo camera: it should be 57 mm, not 273 mm. Table 2 in the draft PDF has been updated accordingly.
  • 2017-05-10: Add the calibration and data sheet of the MTi IMU.
  • 2014-05-15: All images have been rectified again such that both stereo image centers coincide. This simplifies the number of parameters needed to perform stereo SLAM or visual odometry. Calibration files of rectified images have been updated.
  • 2013-10-09: First version online.

Related datasets