Page 112 - DCAP506_ARTIFICIAL_INTELLIGENCE

Artificial Intelligence

Notes  A few broad classes of algorithms have bee used to assist with this:
 Pearls’s message passing method.
 Clique triangulation.
 Stochastic methods.

 Basically they all take advantage of clusters in the network and use their limits on
the influence to constrain the search through net.
 They also ensure that probabilities are updated correctly.

Notes As information is local information can be willingly added and removed with
minimum result on the entire network. ONLY affected nodes require updating.

Example: Here we portray a practical example from research based here in Cardiff.

We have utilized Bayesian Nets in a Computer Vision application. Here we try to portray the
Bayesian reasoning behind the process.
The objective is to execute a task known as data fusion to attain a segmentation — an explanation of
an object (observed from a set of images) detailing its surface properties. In the instance given
here we contract with a simple cube. So the final explanation will hopefully list its edges and its
faces and how they are associated together.
The input to the fusion process is three preprocessing stages that have removed out edge information
and planar surface information from 2D grey scale (monochrome) images and 3D range data.
So from these three pre-processes we have a record of all lines, curved or straight, a list of all line
intersections (two or three line intersections) and a record of all the surface equations extracted
from both image types. We can now construct the network from these lists of features. As
discussed above, we hypothesize regarding extracted surfaces intersecting. For us to assess these
hypotheses we require to have evidence to sustain or contradict them. The evidence that we use
is :

 Straight lines extracted from light image.
 Curves extracted from light image.
 ‘Areas of uncertainty’ extracted from depth map.

The two lines lists are produced as discussed above. The areas of uncertainty are found when we
are trying to locate the surface equations of every surface type. Errors are set up in the depth map
where the mask to locate the common surface shape overlaps two or more surfaces, the error
tends to be enlarged thus, providing us a clue that a surface intersection appears in that general
area. So we are using confirmation from more than one source of data.
We continue by taking each of the surfaces in the surface list and a node is produced to represent
it. We then take a pair of surfaces and try to intersect them. If they are perhaps intersecting then
a ‘feature group’ node is produced referencing the surfaces and associated to the children surface
nodes. This process is repetitive for each pair surfaces that we have removed. We now would
like to connect a conditional probability to each of our new nodes. So we now know the surfaces
that could possibly interact in the object. We now connect a probability to these connections. We
do this by locating the equation of intersection, this will be a three dimensional line for two

106 LOVELY PROFESSIONAL UNIVERSITY