Applications to Safe Human–Robot Interaction - 3D Computer Vision: Efficient Methods and Applications - page 299

Fig. 7.17 Estimation of the instantaneous velocity components x ( top ), y ( middle ), and z ( bottom )
of the reference point in the wrist, obtained using configuration 5. Dark grey lines indicate the
measurements, light grey lines the ground truth ('GT') values
to a significantly increased accuracy of the inferred instantaneous motion informa-
tion.
Given the small baselines of the trinocular camera system, it is a remarkable re-
sult that the measurement error of the velocity along the depth axis is nearly identical
to that of the lateral velocity components. Figure 7.17 depicts the estimated veloc-
ity components of the reference point at the wrist (dark grey curves) and the corre-
sponding ground truth values (label 'GT', light grey curves) for a part of sequence 4,
illustrating the accuracy of the motion information obtained by configuration 5. The
absence of a 'phase shift' between the estimated motion and the ground truth in
Fig. 7.17 confirms that the motion estimation stages described in Sect. 2.3.3.2 are
able to provide information about the instantaneous motion properties of the object.
7.4.6 Comparison with Other Methods
At this point it is instructive to compare the obtained evaluation results with the met-
ric accuracies achieved by other body tracking systems. However, this comparison
is difficult due to the rather special application scenario of the work and also be-
cause no large amount of quantitative evaluation data can be found in the literature.
Often the fraction of images for which tracking results are obtained is determined,
but the metric accuracies are only rarely given.