in front of the first one. When both objects are nearly at
the same distance and the occlusion has disappeared
the analysis states two different separated moving ob-
jects at cycle 108 . Then, the object ’Ot’is passing the
other slower object and starting to get occluded by the
overtaken object on the right lane. The hypothesis for
the occluded and not occluded object has to be ex-
changed (cycle 137). At cycle 234 the hypothesis of an
occluded object has been canceled, because the over-
taking object changed back to the right lane and there-
fore disappeared. The standard deviation for the dis-
tance increases with distance as supposed, but stays
below 226 of the real distance. Distinctly one can recog-
nize the changing of the hypothesis in figures 10-13 at
cycle 137.
Figure 11a and b illustrate the same situation for the
lateral offset of the two objects. The standard deviation
stays below 1% of the real lateral offset.
Ablage y: xs[rot--], real[blau-], PS Dy[gruen-.] * 1000
50 100 150 200 250
Zyklen
Figure 11a Estimated lateral offset of object Ot
Ablage y: xs[rot--], real[blau-], PS. Dy[gruen-.] * 1000
jonni
50 100 150 200 250
Zyklen
Figure 11b Estimated lateral offset of the occluded object
Figures 12a and b show the difference between the two
moving objects concerning distance and lateral offset
which is used for the motion analysis. The solid curve is
the exact difference of the estimated state variables in
comparison with the dotted curve which is the low pass
filtered value, but for the position parameters it is not
-30- i Sh e URP I TV ES mis enin drei 1 e je i
-50F-- Wee uro m cmt rn espe Tu ua. + - 4
10 : : ; WM Eat bd
-20
really necessary to filter contrary to the analysis of
velocities (figure 13).
Diff. Abstand dx: xs[rot--], real(blau-], PS Ddx[gruen-] * 10000
20 +-— $m eel rrr er ere i ———— Hm -
10r- : periere. m E — ce — = — =
50 100 150 200 250
kafi3 30 Zyklen
Figure 12a Difference in distance
Diff.Ablage dy: xs[rot--], real[blau-], PS_Ddy[gruen-.] * 10000
50 100 150 200 250
Zyklen
Figure 12b Difference in lateral offset
Diff.Geschw. dxp: xs(rot], real[blau], PS. Ddxp[gruen] * 1
20 T :
15F- d 5 rd posterum eed rt teen eco re enr es : : a
50 100 150 200 250
kafi4_30 Zyklen
Figure 13 Difference in longitudinal velocity
Currently experimental work is being conducted using
images of real scenes on an image processing system
consisting of a PC-based cluster of 7 Transputers (fig-
ure 14); practical results are expected for the near
future.
