Fast>VP

Fast>VP is a fast and effective tool to detect vanishing points in uncalibrated images of man-made environments and automatically orthorectify the involved planes (please see the examples belows). It is a Matlab implementation of the algorithm described in:

Gilles Simon, Antoine Fond, Marie-Odile Berger. A Simple and Effective Method to Detect Orthogonal Vanishing Points in Uncalibrated Images of Man-Made Environments. Eurographics 2016, May 2016, Lisbon, Portugal.

If you use the software, please cite the paper.

Fast>VP is now replaced by >V<, as fast but much more accurate!

Download

  • Full text manuscript: pdf
  • Linux + Matlab source code: zip

Quick Start

  1. in Matlab, run the file « lsd-1.5/compile.m »: an example image should be opened in which line segments are detected,
  2. run the file « run.m » in the main folder and select the folder called « test »: all images in this folder must be processed.

Results are saved in a subfolder of the folder « test », whose name is based on the date & time.

Examples

VP detection Orthorectified images
test3 test3_rec1 test3_rec2
↑ This is a typical example where the so-called Manhattan frame is found. Line segments that contributed to the x, y and z-directions are drawn in red, green, blue, respectively. The horizon line is shown in solid yellow line.  The camera focal length is computed and the involved planes are automatically rectified to make them appear as if observed from a fronto-parallel view.
 test13  test13_rec1  test13_rec2  test13_rec3
↑ Again, a Manhattan frame is detected, but here one horizontal vanishing point is inside the image. We get three rectified images, two for the green vanishing point (one for the buildings on the left and one for the buildings on the right), and one for the red vanishing point.
 test1  test1_rec1  test1_rec2 test1_rec3
↑ The Manhattan frame is detected, as well as a third horizontal direction, shown in brown. This is an example of an Atlanta world.
 test8 test8_rec1  test8_rec2  test8_rec3
↑ No Manhattan frame is detected, but the focal length can still be computed as the zenith is a finite VP and at least one horizontal VP is finite. Here, three horizontal VPs are detected.
 test6 test6_rec1  test6_rec2
↑ The zenith is at infinity, which precludes computing the focal length (except if we assume that the horizontal VPs are orthogonal, but this is unknown a priori). In that case, the horizon line is shown in dashed line. Two horizontal VPs are found and the related planes can still be rectified. It must be noticed that, in that case, the aspect ratios of the buildings cannot be determined.
 test4  test4_rec1
↑ Only one horizontal VP is obtained, and this VP is at infinity (near-fronto-parallel view). Neither the focal length nor the horizon line can be computed, but the vertical plane can still be rectified, up to an aspect ratio.

>V