measuring distance with axis camera

[windtakers]

I've read many different post on measuring the distance of you robot to the hoop and different pdfs like this, but their are a few things I don't understand.

1) How do you account for the rectangle distortion when you are not directly in front of it and at a downward angle.

2)how accurate can I get the distance measurement

3)what is the most accurate way to measure distance with a camera or multiple cameras

4)what would the easiest way to measure distance with a camera.

[JewishDan18]

Originally Posted by windtakers

I've read many different post on measuring the distance of you robot to the hoop and different pdfs like this, but their are a few things I don't understand.

1) How do you account for the rectangle distortion when you are not directly in front of it and at a downward angle.

2)how accurate can I get the distance measurement

3)what is the most accurate way to measure distance with a camera or multiple cameras

4)what would the easiest way to measure distance with a camera.

I'm using Java, but I should be able to help a bit.

1) You can account for the downward angle, since you know how high the target is. The perspective transformation is much harder, but I am having pretty good results by ignoring it for the most part.

2) I've done very limited testing, but I had a quarter inch of error from 5 feet away. Will get worse when it is further away, but I don't know how much worse yet.

3) I am using the pixel height of the bounding box around the target, and lots of trigonometry to figure it out. If you are willing to use multiple cameras, you can figure more out about the perspective tranform, and thus get a more accurate reading

4) Most likely the way I am doing it, but I would like to stand corrected

[Greg McKaskle]

The size estimate is not too bad even if you ignore the distortion. A better approach, though a bit more expensive would be to use the bounding box to inspect the original image with edge detection methods. You could probably go so far as to find all four corners. But again, once you have a good pixel size for something in the image that you know the physical size, you are able to solve for distance.

As for using the tape width, that is the smallest known element in the image. That will degrade with distance faster than using a larger known size. But as mentioned, it will work provided you are pretty close.

Finally, you can estimate distance several different ways and compare them for accuracy, perhaps even using averages or other techniques for producing a better value than a single technique.

Greg McKaskle