I've been participating in nasa master challenge contest on topcoder (https://www.iarpa.gov/challenges/3dchallenge.html).
Posted source code of my solution on github (https://github.com/krunt). Though I got zero score,
this was a really fun project for me. It's a complex research problem, that requested a lot of free time for experimenting. This is shown in competion results - only 9 contestants in the world got nonzero score. Competion started in august and ended yesterday.
At first I read source code of MVE and Ames Stereo Pipeline. Experimented a lot of with ames stereo pipeline but it didn't get sufficient score (completeness >0.3), so I decided to write my own solution.
After researching a bunch of scientific papers on this subject, found a common way to reconstruct 3d position in multiple images by walking from object space. So I based my solution on these wonderful
phd thesis papers:
[1] Automatic Digital Surface Model (DSM) Generation from Linear Array Images
[2] Baltsavias, E. P. (1991). Multiphoto Geometrically Constrained Matching.
I didn't get much free time to improve solution (having full-time softdev job), haven't implemented refinement with least squares matching from [1], so resulting solution is inaccurate but smooth (used probabalistic relaxation)
Posted source code of my solution on github (https://github.com/krunt). Though I got zero score,
this was a really fun project for me. It's a complex research problem, that requested a lot of free time for experimenting. This is shown in competion results - only 9 contestants in the world got nonzero score. Competion started in august and ended yesterday.
At first I read source code of MVE and Ames Stereo Pipeline. Experimented a lot of with ames stereo pipeline but it didn't get sufficient score (completeness >0.3), so I decided to write my own solution.
After researching a bunch of scientific papers on this subject, found a common way to reconstruct 3d position in multiple images by walking from object space. So I based my solution on these wonderful
phd thesis papers:
[1] Automatic Digital Surface Model (DSM) Generation from Linear Array Images
[2] Baltsavias, E. P. (1991). Multiphoto Geometrically Constrained Matching.
I didn't get much free time to improve solution (having full-time softdev job), haven't implemented refinement with least squares matching from [1], so resulting solution is inaccurate but smooth (used probabalistic relaxation)
No comments:
Post a Comment