Results and Discussion
the analog to digital conversion stage inside the sensor . The centroid calculation de- termines that pixels below a certain threshold are noise and discards them. Testing has shown that using a pixel value of 4 as a threshold in the centroid calculation works well.
The third source of noise is due to the noise on the LED drivers. Ripple from the mains
supply introduces a 50 Hz oscillation in pixel intensity.
s is shown later this does not
affect markers with large intensities as much as markers with lower intensities.
There are also unknowns in the pose estimation stage.
gain number precision is a source
of noise although negligible. If the correspondence algorithm links a centroid to the incor- rect marker then this could introduce a large uncertainty, however, this is a problem that can be removed entirely by the successful application of the correspondence algorithm. The last unknown relates to uncertainties that could be introduced by the pose algorithm itself. In the algorithm described in Section 7.4, the minimisation terminates when the cloud size is below a specified value or the step size is reduced to below a specified pre-set. Setting the required cloud size directly affects the maximum performance that could be expected with this algorithm. Ideally the cloud size should be set low enough so that the uncertainties due to the pose algorithm are lower than the uncertainties introduced by the
variation in centroid positions.
Black Spot module testing (2D testing)
In this section the Black Spot module test results are presented. By definition the per- formance of the markers are also tested as testing can only occur with both marker and camera. This section discusses tests up to Point 1 in Figure 8.1. The main output from the modules is a stream of centroids. The variation in the centroid positions is measured while the Black Spot remains stationary. This variation is used to assess how the uncertainties introduced in Section 8.1 affect the performance. The approach taken is to isolate each variable and determine what effect it has on the overall centroid variation. It is assumed that the variables are linearly independent allowing each unknown to be treated in isola- tion. The parameters that can easily be controlled are the marker intensity, the observed
marker image size, and the marker image position in the FOV. camera and marker poses.
ll three depend on the
By fixing the poses, the effect of marker intensity on centroid variation can be determined (Section 8.2.2). Marker image size is isolated in Section 8.2.4. The marker image position is tested in Section 8.2.3. In Section 8.2.5, the variation in centroid position over a large period is tested. This tests not only the module but the entire test setup as any expansion or contraction due to variations in temperature of the equipment will cause the centroid positions to move. The performance between modules is tested in Section 8.2.6.