Selecting control points for inverse perspective mapping (using fitgeotrans)

Hello everyone,

I'm starting out with image processing and I have an image from camera and need to change it to bird's eye view using fitgeotrans function. Given that I only have one image, how do I find out the moving and fixed points? I was also unable to understand the concept of these control points based on the matlab documentation using the rotated checkerboard image. I hope someone can help me to understand generally understand the concept or offer me help with my specific problem, which is determining control points for bird's eye view transformation. An example of the canera image is attached below.

[Matt J]

The image looks bird's eye view already. Aren't we looking down on a road from above?

What would a bird see differently in this picture?

[Thana Seelaraj Baiskaran]

this image is a cropped image taken from a camera at a height where the raspberry pi camera will later be. It is not a bird's eye view image yet.

@Matt J could you explain how to choose the fixed and moving points based on the image below.

[Thana Seelaraj Baiskaran]

assuming the lane (as shown in the second image, first image might not be the best example) is a straight lane, without any curve, a bird's eye view will show a lane made of two parallel lines.

Answer 1

[Matt J]

Given that I only have one image, how do I find out the moving and fixed points?

So, notice that you no longer have only 1 image. You have the photograph and you have the conceptual drawing of the bird's eye view that you provided in a later comment. You just need to pick corresponding points from both, perhaps the 4 corners of the road or the road markings.

In any case, a perspective mapping will not work since straightness of lines is not preserved (the lane markings are curved in your first image but not the second). Perhaps the polynomial transformType could be attempted.

[Thana Seelaraj Baiskaran]

If i'm determining the control points using an image (will be taken using camera later on) of a straight lane, I will be using four corners of the road from image. For now i will use the conceptual drawing (with some measurements) as the reference. Assuming i know the size of the image and X,Y coordinates of the four points of the road, is my fixed points [Ax Ay;Bx By; Ax Cy; Bx Dy ] and moving points [Ax Ay;Bx By; Cx Cy; Dx Dy] ?

[Matt J]

Yes, that seems natural to me. If your pixelization is fine enough, you could even treat the road markings as thin rectangles. Taking the 4 corners of each road marking, you could have a total of 8 control points per image.