EFFICIENT LINE EXTRACTION FOR DIGITAL ARCHIVES OF CULTURAL HERITAGE 2.1 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
USING OPTICAL FLOW AND TRIFOCAL TENSOR Lit 
thr 
Y. Kunii*, H. Chikatsu Tis 
int 
Tokyo Denki University, Department of Civil and Environmental Engineering, ff 
Hatoyama, Saitama, 350-0394, Japan 
(kunii, chikatsu)@g.dendai.ac.jp i 
Commission V, WG V/4 rl 
KEY WORDS: Cultural Heritage, Extraction, Sequences, Matching, Modelling, Visualization d 
h, 
ABSTRACT: x 
In order to perform object modelling by image processing procedures, line or feature extraction and stereo matching used to perform. 00 
However, there are some issues for efficient 3D modelling of objects. In particular, efficient line matching for reconstruction of Th 
objects is needed to be resolved. With this objective, a robust line matching method which includes optical flow and trifocal tensor | Fur 
has been investigated. This paper focuses on more efficient line matching using Least-Median Squares (LMedS) and automatic 3D | by 
modelling of historical structure. | seq 
line 
1. INTRODUCTION 2. LINE MATCHING BY OPTICAL FLOW | 
Recently, efficient spatial data acquisition and visualization Image sequences for the Koma house were taken while amateur | 
have been receiving more attention from the view point of video camera was moving in horizontal direction by hand held, | 
construction of digital city, VR museum including digital and line matching was performed by following procedures in | 
archives. Generally, in order to perform object modelling Figure 1. Detail procedures of the line matching method are as | 
through digital images, line or feature extraction and stereo follows. | 
matching are performed, and various matching methods such as | 
area based matching, future based matching have been proposed. | 
In particular, line often gives important information for object | 
extraction, and accurate 3D results depend on rigorous line 
extraction and matching. 
With this motive, the authors have been concentrating on 
developing an efficient line matching procedure for 3D E 0 000.0 00000 0 00000 5 ous + vafounscansennaseafoounne 
modelling of historical structure using image sequences by V 
video camera. The 3D modelling method contains line matching e tete tentem ER 
which is performed by line extraction and line tracking. The V 
line extraction was performed by Canny operator (Canny, 1986), | ———— coprssosaneenn ousneccansacce 
and line matching was performed using optical flow and trifocal V 
tensor (Beardsley, et al., 1 996). Therefore, availability of image | 0 00 009000990 00090 00000000000 0900 000 pen 00000000000000000000600 
sequences for object modelling was demonstrated (Kunii and M 
Chikatsu, 2003). However, there are still some issues for 3D | - I I sochoscanesoneefeneernneseansaete 
modelling of historical structure such as acquisition of image x 
sequences or automatic line extraction for the first image, and | ® 0.00.0.000090.00.0000000 0000080000 a enne 
these issues are needed to be resolved for more convenient and 
efficient 3D modelling. 
In these circumstances, this paper focuses on more efficient line 
matching method using Least-Median Squares (LMedS). The 
LMedS have ability to calculate the trifocal tensor more 
accurately than least squares method, and more accurate line 00000000000 2.2 
matching can be performed. In addition, the LMedS is able to 
remove inaccurate matched lines during the line matching Figure 1. System flow of line matching In o 
procedure. line 
This paper describes more efficient line matching method using esti 
LMedS, and investigates an adaptability of this system to 3D (Lu 
modelling of historical structures “Koma house” which was proc 
built in 17th century (about 300 years ago), and designated as Luc 
national important cultural assets in 1971.