5. Istanbul 2004 
Semi automatic 
1al Archives of 
112-17 
Iman, G., 1999. 
1g CSG Models. 
urement, Object 
e and industry, 
2. 
ymputations, 3rd 
P.J., Bunke, H., 
zgibbon, A., and 
n of range image 
ns on Pattern 
689. 
POB Magazine). 
/01033d-laser- 
1ree-dimensional 
00. Progress in 
range image 
ie coH Pattern 
647. 
raison,P., 1980. 
roc. of Autofact 
overing quadrics 
veys, 34(2). pp. 
^ Flannery.B.P.. 
rt of Scientific 
ress, Cambridge, 
thods For Fitting 
parison. The 7th 
er Graphics And 
JSA. 
:ling: a historical 
, Comput. Graph. 
with quaternion 
nce on Computer 
'mented Reality. 
(Accessed Nov 
Heuvel van den, 
ry for Reverse 
yuter-Aided Civil 
4-274 
A PHOTOGRAMMETRIC METHOD FOR ENHANCING THE DETECTION OF BONE 
FRAGMENTS AND OTHER HAZARD MATERIALS IN CHICKEN FILETS 
S. Barnea^' V. Alchanatis" H. Stern? 
* Dept. of Industrial Engineering and Management, Ben Gurion University of the Negev, 84105 Beer-Sheva Israel — 
(barneas, helman)@bgumail.bgu.ac.il 
? Inst. of Agricultural Engineering, The Volcani Center - ARO, 50250 Bet Dagan, Israel — victor@volcani.agri.gov.il 
Commission TS, WG V/I 
KEY WORDS: Close Range, DEM, Automation, Color, Industry, Photogrammetry 
ABSTRACT: 
The research suggests a system that automatically generates Digital Elevation Model (DEM) of a chicken filet. The 
DEM is generated by close range photogrammetry methods, using stereoscopic images of the chicken filet. A major 
problem of the photogrammetric process is the uniformity of the chicken filet’s color and texture, which makes it hard 
to find matching-points in the two stereo images. The research suggests a method to enhance the accuracy of the 
photogrammetric process, based on the projection of a continuous multi-color pattern on the chicken filet. The 
projected pattern is iteratively designed based on the object's optical and textural features and on the image acquisition 
and projection systems color definition abilities. System measurement accuracies were within the specified limit of .5 
mm in height. 
1. INTRODUCTION 
Accurate detection of bone fragments and other hazards in de- 
boned poultry meat is important to ensure food quality and 
safety for consumers. Automatic machine vision detection can 
potentially reduce the need for intensive manual inspection on 
processing lines. 
Current X-ray technology has the potential to succeed with low 
false-detection errors. X-ray energy reaching the image 
detector varies with uneven meat thickness. Differences in X- 
ray absorption due to meat unevenness inevitably produce false 
patterns in X-ray images and make it hard to distinguish 
between hazardous inclusions and normal meat patterns. 
Although methods of local processing of image intensity can 
be used, varying meat thickness remains a major limitation for 
detecting hazardous materials by processing X-ray images 
alone. 
An approach to overcome the aforementioned difficulties is to 
use an X-ray imager in conjunction with the chicken filet 
thickness algorithm, yielding a thickness-invariant pattern 
recognition system 
2. PROBLEM DEFINITION AND OBJECTIVE 
The present work addresses the problem of automatically 
generating a 3D model of a chicken filet when it passes under 
an imaging system on a conveyer. The 3D model of the 
chicken filet is expressed in terms of a digital elevation model 
(DEM). The DEM should satisfy the following specifications. 
1. The DEM should represent an area of 300 
square millimeters. This is the area 
represented by the X-ray imaging system. 
to 
Every number of data in the DEM matrix 
will represent 1 mm? in world coordinates, 
which means that the DEM matrix 
dimensions will be 300 by 300. This 
resolution is satisfactory for detection of 
most hazardous materials. 
3. The DEM error at each point should be less 
then 0.5mm in height. 
4. The speed of the conveyor is 1 feet 
(304.8mm)/sec, which means that the time 
required generating a single DEM must be 
less then a second. 
3. METHODS FOR GENERATING A DEM 
3.1 Moiré Pattern 
A powerful way of describing a three-dimensional shape is to 
draw contour lines. A Moiré pattern is an interference pattern 
resulting from two superimposed gratings. The geometry of the 
moiré fringes is determined by the spacing and orientation of 
the grids. If we know the grids and can image the moire pattern 
formed on the surface, we can determine the topography of the 
surface. 
There are number of ways to produce moiré patterns and to use 
them to determine object topography. In the shadow moiré 
technique, a reference grid is placed close to the object and 
illuminated so that it casts a shadow on the object. The shadow 
of the reference grid is distorted by the object's topography. If 
the light source and viewing point are at the same distance 
from the reference grid, the moiré fringes represent contours of 
depth with respect to the reference grid.