template matching [44–46] to match a marker image against a potential marker. This would be less intensive than template matching during segmentation because template matching would only be performed on each ROI rather than the entire frame. By calculating the size of the marker image, the approximate distance to the marker can be calculated and this could be used to decide whether the marker is too close or too far away to be used.
Duplicate ROI detection
n algorithm to detect whether identified ROIs are duplicates would be useful. In some cases it has been observed that rapid movement causes two ROIs to merge. This is move- ment greater than the maximum sweep velocity found in Chapter 3. This can be caused by a ROI containing part of an adjacent marker due to rapid movement. The centroid algo- rithm is then biased by the partially visible marker. The outcome is that the ROI slides on top of the other ROI and duplicates it. Periodically comparing the coordinates of known ROIs could provide a way to eliminate this problem.
nother useful algorithm would be one that assesses the quality of the markers based on
measured properties of their images.
Quality, in this context, is defined in terms of the
variation in centroid positions while the Black Spot module is stationary.
marker is one with a low centroid variation. Knowing the marker quality would aid 3D pose estimation since poor markers could be ignored or given a lower weighting. The quality metric could be a scalar equal to the weighted sum of a number of measured pa- rameters. For example, these parameters could be the maximum pixel intensity, the extent
of image saturation, and the size of the marker.
The peak pixel intensity is a useful measure as it has been shown that the location precision is approximately doubled for a doubling in peak intensity . If the image is saturated then the maximum intensity cannot be represented by the number system. Shortis et al. 1994  note that in their simulations, the location accuracy of the centroid calculation decreases with an increase in saturation of the image. The marker image size also has an affect on location precision. Clarke et al. 1993  note that little improvement in precision
is obtained by increasing the marker size above 5 pixels in diameter.
lso, the position of
the marker image in the ROI may be related to the quality of the centroid. For example, if the marker is not completely contained by the ROI then the centroid will not be comparable with the centroid of the full marker.