183
their measuring precision is to be estimated to O,,,, - 0.15 mm (=point cloud with 30-40 matched 3D points per single
finite element mesh) and Geo - 0.11 mm (=profiles measured with 0.1 pixel parallax precision). The empirical stan-
dard deviation of the measurements derived from the DCS 200 stereo pair were calculated from r.m.s. values which
were obtained with the help of the reference data. In case of the point cloud measurements the r.m.s. values were de-
termined by interpolating the reference data into the facets of the point cloud. The r.m.s. values of the profile meas-
urements were directly calculated from coordinate differences. Table 3 shows the compilation of the theoretical ( 7
O,,,) and empirical standard deviations ( = r.m.s.) for both measurement options. Additionally, table 3 includes the
internal precision estimation 0, Of the point cloud based on the sigma naught and the normal equation matrix. À cor-
responding value for the profile measurements is not included in table 3, since the single points of the profile meas-
urements were only intersected by two rays and thus no reliable internal precision estimation was possible.
The principal goal of the accuracy evaluation was to check the expected internal measuring precision of the two meth-
ods by independently measured reference data. Since the absolute orientation of both stereo pairs was carried out with
the same control points but with different pixel sizes, the reference data and the DCS 200 measurements might have
systematic discrepancies. However, no significant bias values were obtained in the r.m.s. value calculation. The val-
ues in table 3 indicate for both measurement options a good correspondence between the theoretical standard devia-
tion O,,, and the empirical standard deviation r.m.s. . Also, the internal precision estimate G, , for point cloud meas-
urements tends to be in the same order of the magnitude like the corresponding theoretical and empirical standard
deviations. In comparison, the profile measurements appear about 25 % more accurate, which has also been pointed
out in 3.3.3. If the precision values are propagated to a full scale model, a measuring precision between 0.75 mm and
1 mm is to be expected. Altogether, the figures in table 3 confirm the expected values and demonstrate the successful
applicability of the calibrated Kodak DCS 200 still video cameras.
CAL1 1 CAL1_ CAL1 Option # check o o. r.m.s.
1251 points thes int
1 1 point cloud 17500 017 0.14 0.16
6661 profiles 1557 0.11 - 012
0. ;
27 27 O neo ^ theoretical standard deviation
0.001 0. O , : theoretical standard deviation based on sigma naught and
0. 0. Q
0. 0.041 0. r.m.s. : empirical standard deviation
0 0.001 0.
0.1151 0116 0116" Table 3: Theoretical and empirical standard deviations
lef lef let for DCS 200 surface measurements (units
are mm)
notsignif| not signif| not signif ;
not signif| not signif. not signif object #check | go Om r.m.s.
d points
me for m 2 14 28 int. door 7968 0.10 0.13 0.11
ent
me for calibration ; S i4 O quo Oi 8: see alsotable 3
Table 1: Different calibration runs (units for coordi- Table 4: Theoretical and empirical standard deviations
nates and focal length are mm) for point cloud measurements on an interior
door model (units are mm)
CAL1 CAL2
Xo(0xo) -0.042 (0.0004) 0.002 (0.0007)
yo(cyo) 0.040 (0.0006) 0.137 (0.0010)
c(o.) 27.062 (0.0008) 27.201 (0.0013)
aca) | 0.116*10° (0.0002*10°) | -0.124*10° (0.0004*10*)
ro 6.0 6.0
Ox, Oy 0.0007 0.0013
Table 2: Calibration results for the two test cameras
IAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop “From Pixels to Sequences”, Zurich, March 22-24 1995