The project called for a vision based robot. Achieving object level recognition in vision would be large project in itself. For our purposes, we needed pose determination and target identification through vision. We tried using different colored strips like in a zigzag manner so that the orientation can be determined on the basis of ratio of the colors seen and the location based on the depth of the strips in the image. It turned out that the camera we were using could not resolve edges well at the distances we wanted. And without a sharp edge, the algorithm gave inaccurate results (or should I say imprecise?).

Keep it simple, stupid – the KISS Principle

We came up with a simple method of guiding the robot. Just one strip placed on the floor, if the camera is perpendicular to it, the image of the strip appears horizontal or else it is oblique. If the distance from the strip is estimated, then the angle through which the robot has to turn can also be estimated.

(please ignore the things in the background - pliars etc. Our bot does it...so u do it to!)

We knew absolute colors didn’t make much sense to detect regions of an image. We tried to put work out in terms of ratios of primary colors. It worked whenever there was reasonable illumination.

But our encounter with computer vision in one brilliant statement by Gowri –

        “Computer vision is all about setting the right thresholds”.

Problem solved? Offcourse, but few more left to handle.

The next bit was when the bot sees a corner- which strip should it be guided by? Obviously the one towards the side it needs to turn. Ok acceptable, but then what if it undershoots and keeps seeing the same side?

Ok, that means its got to learn to differentiate the consecutive sides of its border.

Does that mean use strip of another color? Well, after having taken days to teach it to recognize red- teaching it another color was not our way of doing things. It knows red and it knows strips - use what it already knows : thats our way (of escaping more work!). So the bot differentiated the sides by structure of the feature - number of red strips.

Moral: Use whatever you have to the maximum!