HIGH ACCURATE LOCATION ON DIGITAL IMAGE AND 
APPLICATION IN AUTOMATIC RELATIVE ORIENTATION* 
Prof. Zhang jianging, Prof Zhang Zhuxun, Mr. Wang Zhihong 
Institude of Digital Photogrammetry 
Wuhan Technical University of Surveying and Mapping 
39 Lo-Yu Road, Wuhan, Hubei, 
P.R. China 
Commission III of ISPRS 
ABSTRACT 
A method, positioning accurately the corner and cross points on digital images, is presented in this paper. Based on the 
feature extraction by interest operator, the lines, forming the corner and cross point, are located firstly. Then their 
intersection can be determined. Various simulated images have been used in the test of position accuracy, which is much 
better than other method's one. The algorithm has been applicated in the relative orientation of real digital image pairs and 
the results are quite satisfactory. 
key words: High accuracy, Location, Orientation, Straight line, Corner, Spread function 
1 INTRODUCTION 
Location of point and straight line is the basic step of 
Digital Photogrammetry and Computer Vision. So far, 
many methods {or location of point and straight line have 
been proposed. Some popular methods are introduced 
following : 
1.1 Moment-preserving method 
1.1.1 Gray moment-preserving edge detection 
(Tabatabai, 1981) If g(i,j) is gray value at pixel (1j), then 
the k-order gray moment of digital image is defined as : 
my- I/NZXgk(j)-1/N X Ni gik (1) 
i} 1 
where N is the total number of the pixel in the image, Nj is 
the total number of the pixel in the image with value gj. 
For the one-dimensional case, ideal edge may be 
expressed as: 
I(x)7g1 * (g2-g1) u(x-x0) (2) 
Where g1 is the signal value below the edge, g2 is the 
signal value above the edge, xg is the location of the edge, 
and u(x) is the unit step function. Since there are three 
parameters unknown. the first three moments are chosen 
to solve them: 
m;=(x0-0.5)/N gi ^ (N-x940.5)/N g5l J=1,2,3. 
In particular, the solution for x0 is : 
xp=N/2-(1-c/sqrt(4+c2 ))+0.5 (3) 
where 
c-(3m,m;-m3-2m;3y/ o? 
0?=my-m,2 
1.12 - jon (Liu. 
1990) Hf a corner Py(xp,yp) exists in a circular region with 
radius r. Two intersection points of the boundaries of 
corner and the circumference of the circle are P1(x1.y1) 
and P2(x2.y2). The area of the region between angle 
P1Pop; and arc pip; is: 
A=[ar+x2y,-X1y2)-Hx1-x2)(y2-yo) 
(y1-Y2)(x2-x0) 2 (4) 
mass moments: 
Mx =g2{(x1-x2)(r2-x1x2-ÿ1ÿ2)+(ÿ0+y1+y2) 
[rx yy)» (ry) ax) Ve 
My =82{(91-y2)(r2-x1x2-ÿ152)+(xg+x1+x2) 
Tocex2)yz-yo)ri-y2)(-x9)]Ve 
Mxy-g2(r^(y^-y13-x2x12/16 4 [(xyg- xgy;) 
(xzyz*xpyp) - (xoyrxiyo)(xiy rexoyg) 12 
*xoAyit-y22)yg 1 x23-x42) 24) (5) 
and — xj?ryj2-r? 
xphy ler? (6) 
So the point Py(xy, yp), Pi(x1,y1) and Py(x3,y;) can be 
determined by solving the six equations. The method is 
sensitive to noise because noise can notably effect the 
moments. 
1.2 Wong detection method 
Wong and Wei-Hsin (W ong. 1986) have developed a 
method for the location of circular targets on digital 
images. The first thresholding converts the window area 
to binary image : 
Threshold =(min pixel value + mean pixel value)/2. (7) 
Subsequently the position (x,y) and roundness (r) of the 
target can be computed 
* The investigation has been supported by National Nature Funds and Doctor Funds of P.R. China 
co P ona 
where r 
(p=0,1.- 
origin a 
threshol 
not rout 
Trinder 
variatio 
and the 
the gra) 
under | 
Trinder 
few suc 
1.3 Mik 
Let f(s, 
Consid 
with a 
known 
where 
Suppo: 
of para 
‘Thus le 
of par: 
For or 
equatic 
Using 
calcula 
determ 
With se 
have re 
functio 
1.4 Ho 
Hough 
parame 
also ot 
with t