Mountain Dataset

The VENTURI Mountain Dataset is a collection of 12 outdoor sequences, each of them enriched with carefully verified Ground Truth (GT) data.

All of the sequences were acquired using the same mobile device, a Sony Ericsson XPERIA Arc S, equipped with Android 2.3.4.

Captured data

Each sequence includes the following captured data:

  • The video stream (mp4, 640×480) and the timestamp of its first frame.
  • A timestamped dump of the on-board accelerometer (BMA150).
  • A timestamped dump of the device’s orientation (from SensorManager.getOrientation() ).

Ground Truth

The provided Ground Truth includes:

  • The intrinsic parameters of the camera (the same for all sequences, more details below).
  • For each sequence, its fixed geo-location (latitude, longitude, altitude), i.e. the user was not moving around during acquisition.
  • For each sequence, a list of all geographical toponyms visible from there, with their name, location, distance, angular position, etc.
  • For each sequence, a vectorial representation of all mountain profiles visible from there, with positions provided both in angular and metric coordinates.
  • For each frame, a set of 2D points representing correspondences with analogous points in other frames. Point identities are coherent across the same sequence, i.e. the same real world point has always the same ID in all frames where it appears.
  • For each frame, its absolute orientation, expressed as yaw, pitch, roll angles (in degrees). yaw is null when optical axis points towards North, and grows clockwise, i.e. North=0°, East=90°, South=180°, West=270°. pitch is null when optical axis is horizontal, it is greater than zero when pointing up, less than zero when pointing down. roll represents the rotation of the camera around the optical axis.

Details and data formats

The intrinsic parameters of the camera are the same for all sequences:

  • Image size: 640×480
  • Focal length expressed in horizontal pixels: 578.439
  • Focal length expressed in vertical pixels: 571.025
  • Displacement of the optical center with respect to the central pixel of the image: (Δr,Δc) = (-2.59964,-3.75566)
  • Radial distortion parameters: a=-0.0171494, b=0.0643652, c=-0.0594286.

Each sequence is saved as a directory <seq>_<when> containing the following files:

  • <timestamp>.mp4 : video file, with timestamp of the first frame encoded into the filename.
  • BMA150_accelerometer.log.txt : text file, where each line has 4 values: a timestamp, and 3 measures from the accelerometer.
  • Orientation_getOrientation.log.txt : text file, where each line has 4 values: a timestamp, and 3 orientation angles.
  • <seq>_gt.xml : XML file containing GT data (geo-location, intrinsic, orientation and points for each image, absolute direction of each point). The format is designed to be human readable.
  • <seq>_geo.xml : XML file containing geographical data (toponyms and mountain profiles). The format is designed to be human readable. The reference system for all metric coordinates has its origin exactly at the observer’s position, x-axis points East, y-axis points North, and z-axis points upwards.

Download

Mountain Dataset Download
Title Download Geo-location Notes
F1 Neves Lake 267 frames, 11M (46.94170, 11.78626, 1866m) clouds, dark, stable
F2 Ponte di Ghiaccio 279 frames, 12M (46.94711, 11.73926, 2545m) clouds, people, stable
F3 Sunset over Paganella 226 frames, 9.2M (45.97423, 10.98172, 669m) clouds, roll, sunset, yaw
F4 Green vines 140 frames, 5.7M (46.03631, 10.98150, 428m) blue, green, stable
F5 Ruins 309 frames, 13M (46.06929, 11.15467, 474m) blue, buildings, pano, trees
F6 Monte Misone 526 frames, 22M (45.95541, 10.83291, 1806m) distant, haze, pano
J1 Paul’s window 144 frames, 5.9M (46.06689, 11.15112, 420m) buildings, clouds, shadows, stable
J2 Neves Lake 2 346 frames, 15M (46.94170, 11.78626, 1866m) dark, fast, pitch, roll, yaw
J3 Sky over Stivo 188 frames, 7.7M (46.03580, 10.98223, 430m) backlight, clouds, dark
J4 Cavedine 280 frames, 12M (45.98133, 10.98424, 620m) dark, pano
J5 Cima Ghez 198 frames, 8.1M (46.06928, 11.15469, 474m) buildings, fast, pano
J6 Care Alto 214 frames, 8.7M (45.95541, 10.83291, 1806m) distant, pano

This dataset was created inside the European funded project VENTURI. You are free to use it, but we would appreciate it if you would include a reference to its source in any derived work and publication you write. Thank you.

In case of problems, please contact Dr. Paul Chippendale chippendale@fbk.eu

Acknowledgments

The VENTURI consortium would like to thank the ‘stage’ high-schools students Fabiana Mori and Jacopo Strapparava for their contributions in manually labeling the point correspondences.

Related Publications

Learning Contours for Automatic Annotations of Mountains Pictures on a Smartphone, Lorenzo Porzi, Samuel Rota Bulò, Paolo Valigi, Oswald Lanz and Elisa Ricci, Eighth ACM/IEEE International Conference on Distributed Smart Cameras, 2014 – pdf (1.5M)

Comments are closed.