Rekonstrukcija scene iz več pogledov med vožnjo poljskega robota

magistrsko delo

Jože Kraner (Author), Damjan Zazula (Mentor), Jurij Rakun (Co-mentor)

Abstract

V kmetijstvu se vse pogosteje srečujemo s poljskimi roboti, ki za navigacijo uporabljajo laserski merilnik oddaljenosti in video kamero. V magistrskem delu predstavimo nekaj poskusov s takšnim poljskim robotom. Iz trojic zaporednih slik rekonstruiramo 3D-sceno, določimo oddaljenosti testnih točk in razdalje primerjamo z referenčnimi laserskimi izmerami. Predstavljamo svoj postopek, ki optimalno izbiro korespondenčnih točk doseže s povezavo in primerjavo rekonstruiranih položajev in usmeritev kamere pri izbrani trojici slik. Predlagani kriterij je pri iskanju zanesljivega in usklajenega nabora korespondenčnih točk uspešnejši od znanega postopka na osnovi homografij. Testne točke, rekonstruirane z našim postopkom, od referenčnih izmer odstopajo v povprečju za 2,76 ± 0,76 cm, kar je 42,6 % manj kot rekonstruirane točke na osnovi homografij.

Keywords

3D rekonstrukcija;poljski roboti;računalniški vid;epipolarna geometrija;korespondenčne točke;magistrske naloge;

Data

Language:	Slovenian
Year of publishing:	2017
Typology:	2.09 - Master's Thesis
Organization:	UM FERI - Faculty of Electrical Engineering and Computer Science
Publisher:	J. Kraner
UDC:	004.925.8:007.52(043.2)
COBISS:	20539158
Views:	1090
Downloads:	106
Average score:	0 (0 votes)
Metadata:

Other data

Secondary language:	English
Secondary title:	Structure from motion based on multiple views captured by a field robot
Secondary abstract:	The use of field robots is on the rise in agriculture. To navigate, they rely on the use of distance sensors and video cameras. This thesis presents experiments with such a field robot. We reconstruct the 3D scene from trios of consecutive images, estimate depths of test points, and compare the estimates with referential measurements obtained from the laser distance sensor. Our procedure for optimal selection of corresponding points connects and compares the reconstructed positions and orientations of the camera in the selected trio of images. The proposed criterion of finding a reliable and coherent set of corresponding points is more effective than the known, homography-based method. Test points reconstructed with our procedure deviate from referential measurements for 2.76 ± 0.76 cm, on average, which is 42.6% less than with homography-based reconstructed points.
Secondary keywords:	3D reconstruction;field robot;computer vision;corresponding points;
URN:	URN:SI:UM:
Type (COBISS):	Master's thesis/paper
Thesis comment:	Univ. v Mariboru, Fak. za elektrotehniko, računalništvo in informatiko, Računalništvo in informacijske tehnologije
Pages:	XI, 67 str.
ID:	9593918