ite boundaries 
) not coincide 
> outcome of 
youndaries of 
V-cameras of 
.. Y - vertical 
ints on focal 
>s on the left 
; on the right 
les on the left 
inates on the 
oundaries of 
/e mentioned 
o not appear 
Y) and these 
yards) can be 
id right lines 
id right lines 
> of straight 
of errors in 
cal axes of 
ined on any 
ice point on 
obtained on 
vs, that only 
ear zone are 
eclination a 
S especially 
rdinate of a 
] by simple 
xperimental 
s well as all 
' influences 
ical, and in 
of areas of 
nt of pitch 
> to a Near 
of direct 
nts, which 
system by 
| stereopair 
asurements 
zeneralized 
yarameters, 
average) of 
  
direct measurements of these values on the previous large 
interval of time. Such slowly varying parameters are: 
[ - width of a traffic lane; 
h - height of a stereopair center above a NZ plane; 
B — stereopair base, 
For example, the direct measurement of a roll angle 7 is 
determined by expression 
E y. (i5 — 123 — (4 — ta) 
(t + £a \(t3 mia) (3 t t5X(t =/4) 
  
= 
Using a method of dynamic correction of direct measurements 
we get an expression for a roll angle y 
2} 
er 
? . (Z22 —Z24)-(Z23 7:221) 
  
where h is the result of a filtration of direct measurements of 
h . Here the direct measurements look like: 
at detection of the left stereoline 
p. ia od 
147° 2 
  
(t4 — 6) vG +14)] 
at detection of the right stereoline 
fa tf 
])r y À 
i 
o LT 5 312] 
Thus, the outcome of a roll angle measurement directly depends 
only on distance (in pixels) between the bottom points of the 
images of the detected fragments of marking lines (fast varying 
parameter). Direct measurements of these distances are much 
more accurate, because of a smallness of the appropriate 
distances, than direct measurements (for example, angles of 
declination), constructed with use of coordinates of the top 
points of the images of lines fragments. Last mentioned 
measurements are present implicitly at expression for A, but 
there influence is weak, since / is an outcome of a filtration of 
direct measurements of height on the large set of steps of work 
of a system. 
The vertical coordinate of a point of convergence of the road 
marking lines detected in NZ determination is more complicated 
but could be carried out using the same principles. The 
appropriate expressions with use of a method of dynamic 
correction of direct measurements look like 
er i h(1+v(e +4) mez 221) , (22 + vai) (1) 
1 - B - By(t +4)/2) m, 2 
h(1+y(e, +14) S m_ (23) — 254) Qn t») 
  
  
  
pon B(1+(e, +14 )/2) m, T 2 e 
nie # 3m SD x en: Zn) > (23 Le) 3) 
YaL= h(1+y(e, = ))+ By/2 x Ee zn), (23 tay 
i= h(1+ y(t, +14))-By/2 x m_ (22 réu), (v2; * 22) 5) 
  
/ m 
After vertical coordinate of a point of convergence is 
determined for different pairs marking lines pitch angle value is 
calculated by obvious formula. 
At that, it is taken into account, how close to each other are the 
values of vertical coordinates obtained by the formulas 1—5. 
If the values of these coordinates are close at least for three 
pairs lines, it is considered, that the road marking lines are 
detected correctly. 
Then the average value of vertical coordinate is used to 
determine a pitch angle. And this value of pitch angle is 
considered final. 
If all values determined by the formulas 1—5 differ substantially, 
it is considered, that the road marking lines are detected poorly, 
and their directions are defined not correctly. 
In this case choice of an output value of the pitch angle is made 
based on two values: the angle value obtained on the basis of 
correlation tracking of remote objects and extrapolated value of 
the angle, obtained on the basis of a Kalman filter. 
In particular, if the increments of the pitch angle, calculated on 
different zones, coincide (at least on two zones), the output 
value of the correlation block is considered as the output value. 
Otherwise output value is extrapolated value of the pitch angle. 
A dispersion of an estimation of the pitch angle in this case is 
increased, and, as a result, areas of interest for road marking 
lines searching on the following image shoul to be increased. 
Such approach to a choice of an output value of the pitch angle 
is based on simulation of work of algorithms with use of 
experimental registration. 
4. CONCLUSION 
The carried out simulation with usage of experimental 
registrations has shown reliability of operation of the offered 
method. 
-877—