|
From: | William Krekeler |
Subject: | RE: Help |
Date: | Wed, 23 Nov 2011 14:00:45 +0000 |
From:
address@hidden [mailto:address@hidden On
Behalf Of Spencer M
Spencer, Not knowing the particulars of your problem, or what is causing
you trouble, it will be hard for the list to make good suggestions to help you.
Here is an abstract start for things you might try. You need to figure out how
to reduce the picture content by converting it to a logic structure of pixels
labeled by the likelihood of what object they are. Look at trying different edge filters. Or color detectors
(depending on pictorial content), ie if all the screws, nuts, bolts are against
a dark background. Or you could try some sort of shape signature method that for
each point in the image space defines a shape signature that is then matched to
a database of expected shapes. Many of these methods are going to have trouble
with occlusion and you will need to account for partial objects, ie the head of
a bolt can look like the top of a nut if there are no shadows for context. If
all the objects have similar orientations across images you could define a
single object filter and convolve it with each image to flag likely objects. You
will likely need to use some combinations of the suggestions and I expect that
your algorithm will also require, at least a the testing level a ROC curve to
determine the best operating threshold to avoid false alarms. Good luck William Krekeler |
[Prev in Thread] | Current Thread | [Next in Thread] |