Applications to Safe Human–Robot Interaction - 3D Computer Vision: Efficient Methods and Applications - page 298

Fig. 7.15 Spatio-temporal three-dimensional pose estimation results for configuration 4 using
greyscale images
Fig. 7.16 Spatio-temporal three-dimensional pose estimation results for configuration 5 using
greyscale images
In the context of the envisioned application scenario of safe human-robot interac-
tion, the achieved accuracies of pose and motion estimation are clearly high enough
for the prevention of hazardous situations in the industrial production scenario.
In our Matlab implementation, the computation time of the most complex sys-
tem, configuration 5, amounts to 20 s per image triple. A speed of 3-5 im-
age triples per second for a C++ implementation of the full system on standard
PC hardware can be expected when no hardware-specific optimisation efforts are
made.
Instantaneous motion information can also be obtained by directly computing
the temporal derivatives of the poses obtained with the MOCCD and the ICP al-
gorithms, respectively, without applying a specific motion analysis stage. How-
ever, in comparison with system configuration 5, the standard deviations of the
errors of the velocity components obtained based on the directly computed tem-
poral pose derivatives are higher by a factor of about two for the MOCCD and
three for the ICP algorithm. Our proposed motion analysis techniques thus lead