Page 245 - Introduction to AI Robotics

228

b.
a. 6 Common Sensing Techniques for Reactive Robots

Figure 6.16 Segmentation of a red Coca Cola can: a.) original image and b.) result-
ing red regions. Note that some non-object pixels showed a reddish color.

Fig. 6.16 is the output of a threshold on color alone. If a robot was to move
to the “red” in the image, how would it know which pixel to follow? The
perceptual schema could be instantiated for each red pixel; this is simple,
but it would waste a lot of execution cycles. The perceptual schema could
take the weighted centroid of all the red pixels. In this case, it would be
somewhat about where most people would say the center of the can was.
Or, the perceptual schema could attempt to find the largest region where red
pixels were adjacent to each other, then take the centroid of that region. (The
region is often referred to as a “blob,” and the extraction process is known as
blob analysis.)
Color regions can also be helpful in cluttered environments. Fig. 6.17
shows a Denning mobile robot simulating a search of a collapsed building.
The international orange vest of the workman provides an important cue.
The robot can signal a teleoperator when it sees bright colors.

6.6.4 Color histogramming

Thresholding works well for objects which consist of one color or one dom-
COLOR inant color. A different technique popularized by Mike Swain, called color
HISTOGRAMMING histogramming, can be used to identify a region with several colors. 137 Essen-
tially, color histogramming is a way to match the proportion of colors in a
region.