Interim: Artificial Eye

Problem:

Successful artificial sight/vision has long been a heavily sought-after goal in many fields of computer science, especially artificial intelligence. Many specialized approaches and subsequent programs have been created in an attempt to mimic what the human eye does naturally, yet none have even come close to achieving the full range of flexibility that comes so naturally to organic beings.

Solution:

In this spirit, the goal of our team is to create a program which can decipher a picture given to it and successfully identify specific objects within it.

Goals for the Program:

We hope that by the end of the challenge, our program will be able to do several things: 1) break a picture down into several color schemes for easier and more exact analysis, 2) filter out background noise in each of the different color schemes, 3) smooth out the edges (lines) of probable objects or areas, 4) find any corners of said objects or areas, 5) use a dynamic form of template matching to identify the objects, 6) do all of the above in 2 and possibly three dimensions, and finally 7) to do all of the above in real-time by means of a video stream from one or more cameras.

Progress to date:

Most of our work until now has been in the areas of research and design. We have researched various attempts and theories in the field of artificial sight/vision that have helped to guide us on the path towards our prototype program. However, as of now, it is only able to complete the first two goals that we set for it reliably and the third and fourth goals unreliably.
For the first goal, the program simply imports a picture of the user’s choice into a picture box on the main form. At that point the user clicks a button which loads the picture into an array pixel by pixel. It then breaks the picture down into five components: 1) Black and white, 2) Grey scale, 3) Red scale, 4) Green scale, and 5) Blue scale.
The second and third goals were essentially merged as we found it made the program much more accurate to process the picture both before and after the lines/edges were smoothed out.
It scans each component (color scale) of the picture and attempts to find any lines that it deems significant, storing them as objects of a class. It then uses these base boundaries as borders around which to smooth and sharpen details. The program then scans the new, clearer versions of the picture against the older versions to check itself for possible errors. When it deems that it has successfully found all of the relevant edges it moves on to the next step.
For the fourth goal, the program finds any sharp turns in the lines that it previously found in the picture, and stores their locations in another class of objects.
We found that for the second/third and fourth goals, the routines we have written are not robust enough, and only work well for very clearly defined shapes, such as black and white pictures. The program seems to get confused when comparing the five different color scales of the picture and cannot decide on the final lines to follow, giving vastly inaccurate results.

Resources:

ARTIFICIAL INTELLIGENCE
a revision of COMPUTERS THAT THINK?
Margaret O. Hyde

Artificial Intelligence
UNDERSTANDING COMPUTERS
The Editors of TIME-LIFE BOOKS

BRAIN MAKERS
David H. Freedman

MIND MATTERS
James P. Hogan

THE LOGIC OF INDUCTION
Halina Mortimer

Team Members:

Wesley Smalls
Nick Longenbaugh

Sponsoring Teacher: Jim Mims

Mail the entire Team