Lighting

We begin the orange square segmentation problem by working on the lighting issue. From the latter two robot images it is clear that the glare needs to be reduced.

A quick technique to resolve lighting issues is to reduce the image to edges. Most edge detection techniques use the local neighborhood of pixels in order to generate the edge strength. This local neighborhood has the advantage of reducing global lighting issues. This technique was used in one of our very first tutorials on line following. The technique requires edge detection and thresholding followed by detection of the Center of Gravity to determine the robot direction.

We can clearly see that while this technique does have some promise the edges detected also include edges from the white spots embedded in the black tiles. If you are using a surface that does not have these noise elements then edge detection is a possible way to go.

In the case of our trail following robot there is a problem at the end of the course when the robot turns around. During the turn proceedure the robot will momentarily see the tile edge against a lighter carpet. This boundary appears as a very strong edge which will cause the Center of Gravity measure to veer the robot off course.

Instead, we will try another light adjusting technique ...