Users Manual

In the REST-API, a plane is defined by a normal and a distance. normal is a normalized 3-vector, specifying
the normal of the plane. The normal points away from the camera. distance represents the distance of the plane
from the camera along the normal. Normal and distance can also be interpreted as 𝑎, 𝑏, 𝑐, and 𝑑 components of
the plane equation, respectively:
𝑎𝑥 + 𝑏𝑦 + 𝑐𝑧 + 𝑑 = 0
Note: To avoid privacy issues, the image of the persistently stored base-plane calibration will appear blurred
after rebooting the rc_visard.
AprilTag based base-plane calibration
AprilTag detection (ref. TagDetect, Section 7.3) is used to find AprilTags in the scene and fit a plane through
them. At least three AprilTags must be placed on the base plane so that they are visible in the left and right camera
images. The tags should be placed such that they are spanning a triangle that is as large as possible. The larger
the triangle, the more accurate is the resulting base-plane estimate. Use this method if the base plane is untextured
and no external random dot projector is available. This calibration mode is available via the REST-API and the
Web GUI.
Stereo based base-plane calibration
The 3D point cloud computed by SGM onboard the rc_visard is used to find the plane which is farthest away from
the rc_visard. Therefore, the region of interest (ROI) for this method must be set such that only the relevant base
plane is included. The base plane must not be completely occluded by objects. Use this method if the base plane is
well textured or you can make use of a random dot projector to project texture on the base plane. This calibration
mode is available via the REST-API and the Web GUI.
Manual base-plane calibration
The base plane can be set manually if its parameters are known, e.g. from previous calibrations. This calibration
mode is only available via the REST-API and not the Web GUI.
7.5.3 Setting a region of interest
If objects are to be detected only in part of the camera’s field of view, a region of interest (ROI) can be set
accordingly. This ROI is defined as a rectangular part of the left camera image, and can be set via the REST-API
or in the Web GUI. The Web GUI offers an easy-to-use selection tool. Up to 10 ROIs can be set and stored
persistently on the rc_visard. Each ROI must have a unique name to address a specific ROI in the base-plane
calibration or object detection process.
In the REST-API, a 2D ROI is defined by the following values:
• id: Unique name of the region of interest
• offset
_
x, offset
_
y: offset in pixels along the x-axis and y-axis from the top-left corner of the image,
respectively
• width, height: width and height in pixels
7.5.4 Detection of objects
Objects can only be detected after a successful base-plane calibration. It must be ensured that the position and
orientation of the base plane does not change before the detection of objects. Otherwise, the base-plane calibration
must be renewed.
For triggering the object detection, in general, the following information must be provided to the SilhouetteMatch
component:
7.5. SilhouetteMatch 106