4th Year Project "SCAMP"

[Ratfink]

This was my fourth year project it goes through a maze and if it comes across multiple junctions uses the specially designed Vision Chip (contains processor per pixel architecture) to perform image processing to identify which way to turn based on multiple decisions. The car and the project itself i inherited so it was very difficult trying to work around resources already used and the car wasn't ideal for a maze so was definitely a challenge.

I'd just like to thank the Admin and everyone else on the forums for all the help and i hope i can return the favour for someone else.

Anyways here is the videoes link doesn't show everything the car can do our website admin was lazy but shows the maze part :-

http://www.project-scamp.com/testvids.html

[hgordon]

It must have been an interesting challenge to figure out how to employ the potential of the Scamp vision chip, though perhaps there was already a library of image processing functions that you could employ.  I would love to read more about the architecture and instruction set of the Scamp 3, though the links to the P.Dudek technical papers don't seem to be working.

[Ratfink]

Although he has been working on it for over 7 years and constantly improving it every year it isn't made commercial yet. He's trying to pass the chip out to other universities and improving the programming and software debugging enviroment so its still in its development stage.

The instruction set is more like assembly but not quite, There were no image processing libraries all the image processing was done from scratch.

It was very very challenging to try and figure out how to perform image recognition in varying lighting conditions while at the same time create a sign classification algorithm.

We came up with a adaptive thresholding algorithm which is able to select a threshold and vary this across the entire image so the entire image can be seen under many different lighting conditions. The next step was to be able to isolate the shapes and noise so we found that by placing a border around the shape we could effectively look for the border and isolate the sign. After finding a way of clearing all the noise we exploited the shapes characterisitics to build a sign classification program for example a triangle pointing right gave us a linear gradient and depending on the direction its pointing either a negative or positive gradient.

As for links to the papers they were done a while back and i don't think i'd be able to find a working link unfortunately but if you e-mail him he may send them to you.