Fotokaamerate ja tarkvarade võrdlus lähifotogramm-meetriliste 3D mudelite loomisel
Laen...
Kuupäev
2018
Kättesaadav alates
Autorid
Metsoja, Kärt
Ajakirja pealkiri
Ajakirja ISSN
Köite pealkiri
Kirjastaja
Eesti Maaülikool
Abstrakt
Käesolevas bakalaureusetöös käsitlen lähifotogramm-meetria kasutusvõimalusi arheoloogias.
Uurimuse eesmärk oli selgitada, kuidas mõjutab 3D mudeli kvaliteeti tarkvara ja kaamera valik. Võrdlesin kolme tarkvara: 1) Pix4D; 2) Agisoft Photoscan ja 3) Visual SFM koos tiheda punktipilve loomise mooduliga Yasutaka Furukawa's CMVS tool. Erineva hinnaklassi ja tehniliste näitajatega kaameraid oli võrdluses neli 1) Canon EOS 760D; 2) Canon EOS 600D; 3) Canon Ixus 185 ja 4) Sony Xperia Z1 nutitelefoni kaamera.
Tegin kõikide kaameratega täpselt samadest pildistamise asukohtadest fotod kahel erineval pildistamissessioonil (kokku 8 komplekti fotosid). Fotodest erinevate tarkvarade abil loodud esmaste 3D mudelite ehk tihedate punktipilvede välimuse võrdlemisel andis parima tulemuse Agisoft Phtoscan: vaid see programm moodustas kõikide kaameratega tehtud piltidest hea katvusega tihedad punktipilved.
Mudelite geomeetrilist kvaliteeti uurisin programmis Agisoft Photoscan tehtud mudelite puhul. Hüpoteesiks oli, et digitaalsed peegelkaamerad tagavad geomeetriliselt täpsema tulemuse kui kompaktkaamerad. Tulemused olid üllatavad. Esimese pildistamiskorra fotodest saadi suurima geomeetrilise veaga mudel kaameraga Canon EOS 760 D pildistatud fotodest (mudeli viga 5,0 mm) ja täpseim mudel kompaktkaamera Canon Ixus 185 abil (viga 3,4 mm). Teise pildistuskorra mudelite vead olid oluliselt suuremad kui esimese pildistuskorra omad. Suurim viga oli 17,9 mm (Sony Xperia Z1) ja vähim 11,2 mm (Canon EOS 600D). Tulemuste põhjendamine on raske, kuna Agisoft Photoscan on suletud koodiga kommertstarkvara.
Antud töös hindasin mudeli geomeetrilist täpsust vaid 16–23 tahhümeetriliselt mõõdistatud kontrollpunkti suhtes. Kuna arheoloogia seisukohast on oluline uuritava objekti kuju võimalikult täpne edasiandmine, võiks edaspidi võrrelda fotogramm-meetriaprogrammi poolt loodud tihedat punktipilve terrestrilise laserskanneerimise abil saadud referentsandmestikuga.
In current bachelor theses I observe the use of close-range photogrammetry based on phtos taken by consumer cameras in the field of archaeology. Aim of the study is to unveil how the choice of software and cameras influences the quality of 3D models. I compared three different softwares: (1) Pix4D (2) agisoft Phtoscan, and (3) Visual SFM with dense cloud module Yasutaka Furukawa's CMVS tool and four different cameras (1) Canon EOS 760D, (2) Canon EOS 600D; (3) Canon Ixus 185, and (4) Sony Xperia Z1 smartphone camerea D5503. Cameras represent different price segments and tehcnical parameters. I took the photos with all the cameras from identical positions during two photo sessions (obtaining altogether 8 sets of phtos). Comparison of primary 3D models, dense point clouds revealed that in terms of visual appeal, results were the best with Agisoft Phtoscan which was the only software that managed to create visually appealing dense point cloud from photos taken with smartphone camera. Geometric quality of models was futher investigated using Agisoft Pohotoscan software. My hypothesis was that digital single-lens reflex cameras will perform better in terms of model accuracy. The results were surpising. In case of first photo session, Canon EOS 760D images delivered a model with largest total error (5.9 mm) and model made of phtotos taken with Canon Ixus 185 was the most accurate (total error of the model 3.4 mm). Total errors of the modeles built from images of the second photo session were significantly larger. Worst result was obtained using Sony Xperia Z1 camera (total error 17.9 mm), Canon EOS 600D performed best (total error of the model 11.2 mm) It is hard to interpret the results as Agisoft Photosan is a commercial software with closed source code (so-called “black-box” programme). In current research I calculated model accuracy in reference to 19–23 check points measured with total station. In archaeology, it is important to document the exact volumes of structures and artefacts, therefore in the future dense 3D point clouds should be compared to reference data obtained with terrestrial laser scanner.
In current bachelor theses I observe the use of close-range photogrammetry based on phtos taken by consumer cameras in the field of archaeology. Aim of the study is to unveil how the choice of software and cameras influences the quality of 3D models. I compared three different softwares: (1) Pix4D (2) agisoft Phtoscan, and (3) Visual SFM with dense cloud module Yasutaka Furukawa's CMVS tool and four different cameras (1) Canon EOS 760D, (2) Canon EOS 600D; (3) Canon Ixus 185, and (4) Sony Xperia Z1 smartphone camerea D5503. Cameras represent different price segments and tehcnical parameters. I took the photos with all the cameras from identical positions during two photo sessions (obtaining altogether 8 sets of phtos). Comparison of primary 3D models, dense point clouds revealed that in terms of visual appeal, results were the best with Agisoft Phtoscan which was the only software that managed to create visually appealing dense point cloud from photos taken with smartphone camera. Geometric quality of models was futher investigated using Agisoft Pohotoscan software. My hypothesis was that digital single-lens reflex cameras will perform better in terms of model accuracy. The results were surpising. In case of first photo session, Canon EOS 760D images delivered a model with largest total error (5.9 mm) and model made of phtotos taken with Canon Ixus 185 was the most accurate (total error of the model 3.4 mm). Total errors of the modeles built from images of the second photo session were significantly larger. Worst result was obtained using Sony Xperia Z1 camera (total error 17.9 mm), Canon EOS 600D performed best (total error of the model 11.2 mm) It is hard to interpret the results as Agisoft Photosan is a commercial software with closed source code (so-called “black-box” programme). In current research I calculated model accuracy in reference to 19–23 check points measured with total station. In archaeology, it is important to document the exact volumes of structures and artefacts, therefore in the future dense 3D point clouds should be compared to reference data obtained with terrestrial laser scanner.
Kirjeldus
Bakalaureusetöö
Geodeesia ja maakorralduse õppekaval
Märksõnad
bakalaureusetööd, kolmemõõtmeline ruum, modelleerimine, fotokaamerad