Forum

Author Topic: algorithm for matching  (Read 5199 times)

salamander

  • Newbie
  • *
  • Posts: 1
    • View Profile
algorithm for matching
« on: May 27, 2014, 06:33:00 PM »
Hi,

I would like to know how many images the matching algorithm is using for calculating a given point (2 images or all available for the given point), and how these images are selected.

Thanks, Olivier

Marcel

  • Sr. Member
  • ****
  • Posts: 309
    • View Profile
Re: algorithm for matching
« Reply #1 on: May 28, 2014, 07:29:56 AM »
From what I know it uses image pairs only,  but I would also love to have more in depth information (better understanding of the way Photoscan works would hopefully let me make better scans).

On question I have for example: are the points in the Spare Cloud used in the process of generating the Dense Cloud? Or is the Sparse Cloud only used for alignment?

Wishgranter

  • Hero Member
  • *****
  • Posts: 1202
    • View Profile
    • Museum of Historic Buildings
Re: algorithm for matching
« Reply #2 on: May 28, 2014, 09:02:40 AM »
Photoscan is MULTIVIEW, not stereo pairs based alogo.
Sparse Pointcloud is just "cloud" of points-features not the usual PointCloud as you know it....

For deeper info go to http://www.isprs.org/publications/archives.aspx
----------------
www.mhb.sk

David Cockey

  • Full Member
  • ***
  • Posts: 209
    • View Profile
Re: algorithm for matching
« Reply #3 on: May 28, 2014, 12:29:51 PM »
My understanding is that PhotoScan uses two different sets of algorithms.

The first set of algorithms is used for camera alignment, camera calibration and sparse point cloud generation. This set of alogrithms uses multiple photos. Points are identified in each photos; points appearing in multiple photos are identified. For each point all photos containing that point are used in determining camera alignment, camera calibration and sparse point coordinates.

The second set of algorithms is used for dense point cloud generation. Pairs of photos with sufficient overlap which have similar enough orientation are found. The these individual pairs of photos are used to generate sets of dense points using a grid system. Multiple pairs of photos are used with each chunk. The sets of points generated using each pair of photos are merged into a single dense point cloud for each chunk.

David Cockey

  • Full Member
  • ***
  • Posts: 209
    • View Profile
Re: algorithm for matching
« Reply #4 on: May 28, 2014, 12:32:15 PM »
Based on my experience the sparse points are used in selecting pairs of photos to be used for dense point generation.