Page 351 - Introduction to AI Robotics

334
9 Topological Path Planning

Figure 9.7 Derivation of an image signature. Clockwise from the upper left: a raw
image, the image partitioned into 16 sections, and the image with the sections values
shown as crosshatching.

These principles are implicit in the idea of a neighborhood around a dis-
tinctive place. If the robot is in the neighborhood, the views of the landmark
look similar. The main difference is that associative methods use very coarse
computer vision.

9.4.1 Visual homing

Work done by Nelson, 111 and later by Engelson, 50 relied on an image signa-
ture. An image signature is created by partitioning an image into sections.
Fig. 9.7 shows an image partitioned into 16 sections. Next, each section is
examined and a measurement is made of some attribute in the section. Some
possible measurements are the edge density (the number of pixels with edges
divided by the number of pixels in the section), dominant edge orientation
(the angle the majority of the edges are pointing), average intensity, and so
on. The image signature reduces the image to n measurements. One way to
think of this is that it changes the resolution of the image into 16 pixels.
Notice that the image signature in the figure forms a pattern. If the robot