EVEN, Philippe
the area perimeter is a system parameter which can be fixed through the interface. Actually we consider that the operators
should not need to know how this function works. A default value of 5 pixels was assigned. It does not require an accurate
initial matching and affords some discrimination possibilities well appreciated when disturbing segments are close to the
desired one.
This semi-automatic edge segment specification has been evaluated on images with different kinds of difficulties. Image
size was 768x576 pixels and experiments were performed on a Silicon Graphics Indigo workstation. The computation
time depends on the selected segment length and angle. In the worst cases, a latent period of 8 seconds has been observed.
But they are rather rare and in most of the encountered cases, the computation time was around one second. Edge segment
specification tasks are largely speeded up. Their accuracy is comparable with a careful segment specification in manual
mode. We have also evaluated the repercussion of this operation improvement on a whole modelling task. The tests
performed on several types of industrial environments showed a large reduction of the global modelling time.
2 THE TELEROBOTICS CONTEXT
A 3-D geometric model of the remote environment is required at the different stages of a telerobotics mission. In particular
during the execution stage, possible incidents and unforeseen events add specific requirements to the programming and
supervision tasks. Based on the availability of a consistent model, Virtual Reality techniques (Sayers and Paul, 1994,
Natonek et al., 1994, Thibout et al., 1994, Bejczy, 1995, Burdea, 1999) and Augmented Reality techniques (Oyama et al.,
1993, Cannon and Thomas, 1997, Kim, 1999) can provide flexible solutions to enhance the remote control safety and to
reduce the mission execution time.
Of course the model acquisition or update may be performed during some previous inspection step. But it is sometimes
difficult to provide a precise specification of the mission requirements. Moreover the environment may evolve during the
mission. Some robotics tasks may cause unpredictable objects displacements. The availability of on-line modelling tech-
niques using information feedbacks from on-board sensors reduce the model inconsistency problems. The time devoted
to the modelling task must be short compared to the time spared from the model exploitation.
The kind of sensor used for the modelling is selected according to the mission constraints. It depends on the robot payload
capacity and energy autonomy, but also on the environment features. Especially in the case of missions inside radioactive
cells, tolerance to radiation effects, easy remediation and low cost are important criteria. The CCD camera already used
to pilot the robot and to supervise the tasks execution seems quite convenient.
On one hand automatic modelling methods fail to provide the appropriate model, because of the scene complexity, disturb-
ing reflections due to highly specular surfaces, and low contrasts from bad lighting conditions. On the other hand, manual
methods are time consuming. An interactive 3-D modelling principle (fig. 1) has been proposed to meet the telerobotics
field requirements (Even and Marcé, 1988). It relies on a man-machine cooperation through the super-imposition of 3-D
solid primitives over the camera images.
CCD Camera CAD Modeller
Real world N MK Virtual world
Interactive ( Computer |
: C t
Operator |= Modelling —-| Computer
| Consistent 3D model |
Figure 1: The interactive 3-D modelling principle.
The operator brings his structural and functional knowledge on the remote environment. He controls the data collection,
selects the appropriate primitives and matches them on the image details. This mixing of the virtual world with the real
data is a reliable way to ensure the model consistency. The operator settles the final level of detail according to the tasks
to be performed and the available modelling time.
CEA/STR has developed the Pyramide interactive modelling system as an integration platform for the new modelling
methods. Images from an on-board camera are used. The CCD camera parameters are identified during a previous
calibration step. Wide angle lenses are required for piloting or supervision activities. Thus distortion parameters are
also identified and used to remove the distortion effect from the acquired images. This image reconstruction step takes
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B5. Amsterdam 2000. 223