Distributed generation of 3D models of objects from image sets

Rekonstrukcija 3D modela objekta iz nabora slik je zahteven vendar dobro raziskan problem. Potreba po namenski opremi, kot so grafični pospeševalniki, in računska zahtevnost vodita k uporabi superračunalniške infrastrukture. Želeli smo pripraviti postopek, ki bi predstavljal jedro spletne storitve za 3D rekonstrukcijo slikanih predmetov. V sodelovanju s podjetjem Arctur d. o. o. smo preučili postopek rekonstrukcije, izbrali primerno programsko opremo in jo pripravili za izvajanje na superračunalniku Arctur 2. Na različnih naborih slik smo testirali različne vrednosti parametrov v postopku rekonstrukcije, ocenili rezultate in določili optimalne vrednosti posameznih parametrov. Testirali smo na raznovrstnem naboru testnih naborov slik, pridobljenih iz različnih virov, s slikami različnih velikosti in formatov. Tako določen postopek rekonstrukcije proizvede dobre rezultate v večini primerov in je dobra osnova za računsko jedro spletne storitve.Reconstruction of a 3D model from a set of images is a complex but well researched problem. The need for dedicated equipment, such as graphics accelerators, and the computational complexity leads us to the use of supercomputer infrastructure. We wanted to prepare a procedure that would represent a core of a web service for online 3D reconstruction of photographed objects. In cooperation with Arctur d.~o.~o.~we examined the reconstruction process, chose appropriate software, and prepared the selected software for execution on the Arcturnobreakdash-2 supercomputer. We tested different reconstruction parameter values on various image sets, evaluated the results, and determined optimal values for individual parameters. Testing was done on a diverse set of image sets containing images of varied sizes and formats that were acquired from different sources. The established reconstruction procedure gives good results in most cases and serves well as a core of a web service for object reconstruction