4.1. Rigid-Body Transformations
The performed transformations of each static experimental configuration are summarized using the median of the empirical standard deviations
. These spatial mapping errors, obtained by using one calibration, are presented in
Figure 8 for the RPs and SPs, since the applied calibrations and recorded frame rates result in a similar response for both types of generated patterns (refer to
Appendix A.1 and
Appendix A.2). This means that the tested camera exposure times for capturing the calibration images lead to a consistent determination of the calibration parameters under appropriate lighting conditions. For this reason, the adjusted systematic errors of each camera calibration have no significant influence on the determination of the 3D space coordinates; thus, the spatial mapping error remains almost unchanged for the tested static experimental configurations. Contrary to this, the influence of the pattern quality of the listed patterns in
Figure 2 is clearly reflected in the results. From the RPs in
Figure 8A, the highest performance is given by 05 and 06 and the lowest by 01 and 11. From the SPs in
Figure 8B, 02 and 08 show the highest performance, and 05 and 06 show the lowest.
The spatial mapping errors of both types of generated patterns demonstrate that the amount of stochastic information influences the results, since in general the RPs show lower deviations. However, in addition to stochastic information, a good contrast is also needed for the quality of the captured signal to be adequate for the reconstruction and mapping process of the speckle surface. This is the case of the SP 02.
Due to the fact that the image exposure time does not significantly influence the camera calibration and thus the 3D coordinate accuracy, a single camera calibration is used for the dynamic measurements. The spatial mapping errors for the dynamic experimental configurations are presented in
Figure 9 and
Figure 10. It can be seen that the frame rate affects the SPs and RPs differently. While the SP results are influenced more by the image noise than by blurring, the RP results of the 01 and 02 Height are influenced more by blurring. This implies that a good contrast allows better matching results even if there is blurring. However, the 03 Height RP results show a larger effect of image noise than that of blurring.
Comparing the spatial mapping error results of the experiments with and without motion, the error differences between the static and dynamic measurements captured with the same camera configurations are higher with a falling pattern quality. In
Table 3, the median of the
values for the best- and worst-performing patterns are presented. Taking the low-performing patterns 05 and 01, the error of the SP is significantly higher due to the lack of stochastic information of the printed surface. Comparing the high-performing patterns, the error of SP 02 delivers a slightly lower s0 than RP. Due to the difficulty of generating a consistent and proper RP, it can be assumed that in general, a SP has the potential to enhance the quality of the measurements. This gains more relevance for small-scale objects and specific experiments that are intended to be repeated several times.
Using the computed point coordinates corresponding to the dynamic experimental configurations of the best- and worst-performing patterns, the following values are presented in
Figure 11 and
Figure 12 in order to explain their performances: Number of tracked points per time step, vertical cumulative object translation (
), the average distance between the surface points per time step, and the recording time.
The performance of the worst patterns is the result of a fluctuating number of tracked points over time and a larger distance between them. The higher performance of SP 02 compared to RP 06 is due to the higher number of tracked points and the associated smaller distance between them. When analyzing the behavior of the number of tracked points and the average distance between them by experimental configuration, it can be noted that in general the results are noisier with higher drop-weight velocity and image frequency. Furthermore, the number of points fluctuates more than the distance between them. This means that the surface geometry obtained from an experiment is not affected by the frame rate used.
In addition, the
errors of the dynamic transformations show that an increasing frame rate enhances the results, but at an even higher frame rate, the quality of the results can decrease again. From the lowest
values marked in bold in
Table 3, it can be concluded that the object velocity does not significantly influence the coordinate determination, if the exposure time is appropriate. Although the differences between the static and the dynamic bolded values are slightly larger for the RP, the differences are not more than half a
for suitable experimental configurations.
A dependence between object velocity and camera exposure time can be noticed in the results of the dynamical experimental configurations. This means that the higher the velocity, the better the performance at a higher frame rate. This is most evident in the results using the RP 06, where for the first, second, and third velocities, the frame rates 10,000 fps, 25,000 fps, and 50,000 fps perform the best, respectively.
Figure 13 and
Figure 14 illustrate this behavior more clearly with the obtained
from all the transformations for the best-performing patterns.
The influence of exposure time is illustrated in the boxplot distribution of the resulting
values. On the one hand, a larger dispersion of the standard deviations is achieved through long exposure times, which lead to low contrasted images because blurred images present similar gray values of neighboring pixels due to the motion. On the other hand, the resulting
spread at high image frequencies is caused by image noise, since the probability that the light-sensitive part of the pixels detects an object steadily over time decreases with decreasing exposure time. However, the s0 distribution of these experimental configurations shows that performance is more affected by image noise than by blur effects. The effect of image noise can be seen in
Figure 15 and
Figure 16 for the patterns with the best performance under the third dynamic condition, where the
decreases with decreasing exposure time for the tested frame rates. This main tendency is also observed for the SP 02, although the SNR is higher compared to the RP 06.
An attempt is made to find the source of the higher errors with increasing frame rate using the quality dimensions
stereo residual and
intersection deviation given by “GOM Aramis”. No direct correlation is found between the
values and the aforementioned dimensions. It can only be seen in
Figure 17A that the gray value difference between right and left image facets increases with raising frame rate, possibly affecting the image matching performance. The
intersection deviation in
Figure 17B also does not show an influence on the precision of the transformed coordinates but only variable discrepancy between the observation rays calculated from the measurements and the calibration. These discrepancies are the result of random errors that do not show a correlation with the tested exposure times. Similarly, the end velocity computed by the software at different frame rates in
Figure 17C does not seem to be correlated with the
. The measurement stability of the reconstructed geometry for all tested frame rates, shown in
Figure 12, may be the reason for this.