386 
  
  
    
  
  
   
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Specifications OBIECT MOS type SOLID | 
Dimensions : 100(W) X 50(H) X 200(D)mm ex “4 IMAGE SENSOR | 
(not including protruding parts.) | 
Weight : 910 grams LENS 0 | 
Power supply : ACI0O0V 8W | 
Image pick up : Solid state MOS element | 
Image cell number : 320 X 240 ower] A 
Image cell dimension : 27 X27 um MMUNICATION PC-SSOIRASI] — | 
Lens mount  :C standard mount POWER | |... Acio0v | 
Communication port : RS-232C standard interface SUPPLY | 
Functions | 
Image resolution : Fig. Brockdiagram of a Video Tracker | 
320 X 240 total graphic coordinates 256 X 192 effective | 
area. Vertical and horizontal ratio is 1:1. y | 
Object detecting : 4 The Centroid: 
The position of objects is computed from its centroid in A (gx. gy) 1 
binary dimensional data either in bright or black dots 3 Lo] | 
(Fig.3). The threshold level is adjustable. | 
Number of tracking windows : 2 2 | 
Window dimensions : Any rectangular shape | 
Output data : 1 | 
X,Y coordinates, area, frame number and status 
Real-time image processing at 120 Hz. 9 | Fi 
0 1 2 3 4 5 | ca 
x | TI 
2.3 Three dimensional coordinates calculation | 
The Centroid( gx , gy ) i 
Three dimensional coordinates of the objects are calculated | 
from output of ARTIST G140 using Direct Linear gx- (3 (XP(X,y)) gy = 2 (y-p(x,y)) 
Transformation Method without self calibration.(Fig.4) The E(E(P(X,y)) E(E (P(X,y)) 
distortion of normal ITV lens was very small and gave smaller Fig3 Te * osition of objects is computed fom Lis 
Sifert on me acciracy dan die pixel size gave, centroid in binary dimensional data either in bright or 
black dots. 
2.4 Accuracy verification 
Using the computerized XYZ pulse stage, we examined the accuracy of this system in the variation of intersection angle of the 
optical axis of the video trackers and the number of control points. Real coordinates of control points and objective point were 
established by XYZ pulse stage. The minimum pitch of the pulse stage was 2 pm. : 
2.4.1 Intersection angle of the optical axis 
We evaluated the effect of the intersection angles of the optical axis with the following conditions. Fi 
Working distance : 500mm | re: 
Variation of the intersection angle of the optical axis : 30, 45, 60, 90 degrees(Fig.5) | 
Control points : 6 points at the apex of the cube ( 20.0 X 20.0 X 20.0mm) | 
Objective points : 729 points aligned in the cube (20.0 X 20.0 X 20.0mm) at intervals of 2.5 mm 
Residuals of 729 points' coordinates from their real coordinates were calculated. Results shows in Fig.6 When the intersection 
angle was 60 degrees, mean residual was minimum. m 
2.4.2 Number of control points 
The effect of the number of control points was evaluated. 6 or 8 control points were set up at the position showed in F ig.7. 
Objective points : 9261 points aligned in the 20X 20X 20mm cube at intervals of 1.0 mm. The intersection angle was 60 T 
degrees. Residuals of these objective points were calculated(Table 1). In 6 control points this system was more accurate than in 8 
control points, but the difference was very small and we thought that the difference resulted from the coarse pixel of the video 
tracker(effective image resolution: 256x192). 59 % of 9261 points had errors less than 0.1mm, 92% had errors less than 0.15mm. 
Therefore, the accuracy of this system was enough to measure lip movements. | -— 
Tal 
IAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop “From Pixels to Sequences”, Zurich, March 22-24 1995 | /