pattern T is updated. The vectors of unknown input 
pattern T are converted by multiplying the coefficients 
which are determined in the above normalization. 
where a,and b; are the characteristic vector of the frame 
No.i in pattern T and the characteristic vector of the frame 
No.j in pattern R, respectively. This time both / and J are 
20. Each time a new frame is taken, unknown input 
direction are 
e middle point 
(3) 
  
  
  
   
  
  
  
  
  
  
  
  
  
  
         
  
    
  
  
1 1 
(4) y 08| 3 0.8 
= © 
»sponds to the S 0.6} T 0.6 
a, we treated 2 > 
| the relative 1530.4 I 0.4 
angle 6, is 3 E 
aight line tied = >= 0.2 
S = 
(5) 9 © 20:2 
©) 2-04: € -0.4 
a = 
8-0.6 | 9 -0.6 
ment velocity 2 a 
, respectively. =~0.80 5 —0— x — Vy = -0.8 —o— Vx —L— Vy 
d as —+ yz = Vt —0— Vz 3c Vt 
-1 -1 ZX 
(7) 1.13..5. 7.9. 11| 19; db. hd lupa 5 7 9 11 13 15 17 19 
8) Time [frame number] Time [frame number] 
S (a) no action (b) nod 
Z,, and Ó,in 1 
of each action 
leristic vector 0.8 + 
0.5.1 
ws.: Oneof 
| relationship 0.4. i 
ment velocity 
every action. 0.2. + 
sristic vector Pius iO IE 
hand. Each 08 EDS 
he length of = 
leration of the -0.2 
> required for 
rences in the 
normalization 
ocity of which 
attern data is 
-0.4 
-0.6 
  
Displacement velocity [relative value] 
Displacement velocity [relative value] 
    
  
  
  
  
a -0.8 | —o— Vx —— W -0.8 —o- x —&— Wy 
Figure 6 (a)- | —0— Vz = Vt ! —-yz X Vi 
$4£49. 155.07 099 4,513! 450 17:719 KB 5 7 9 3-33::15i 11-19 
ilized here as Time [frame number ] Time [frame number] 
  
sen unknown 
The unknown 
tterns, R, are 
(c) shake (d) tilt 
Figure 6 Templates of head motions. 
439