speed estimate coordinate

[Yeodohoon]

I am practicing with this example from https://github.com/roboflow/supervision/blob/develop/examples/speed_estimation/ultralytics_example.py. In the example code, there is a section that calculates coordinates C and D with the line SOURCE = np.array([[1252, 787], [2298, 803], [5039, 2159], [-550, 2159]]). I would like to understand in detail how the x-coordinates -550 and 5039 are calculated.

[LinasKo]

Hi @Yeodohoon 👋

Are you following this guide? https://blog.roboflow.com/estimate-speed-computer-vision/

If not, check it out - it may help clarify the approach.
Let me know if it doesn't help 😉

[Yeodohoon]

Hello @LinasKo
I watched all the guides and YouTube explanations.
When calculating the coordinates of points C and D in this part of "Math Behind Perspective Transformation", I don't understand how to find the x coordinate.

[SkalskiP]

Hi @Yeodohoon 👋🏻 In a case like the one described in the tutorial, where not all points lie on the image, I usually determine the zone geometry by trial and error.

I start by drawing a simple polygon in Colab, such as a 100x100 square. A good idea is to print the image using matplotlib. This displays a scale from which we can estimate the coordinates of the vertices of our polygon.

Looking at the vertical axis, I conclude that I want to cut off the road around y = 750. From the x-axis, I roughly estimate where points A and B should be. The remaining two points will be on the bottom edge of the image. I know the image has a resolution of (2160, 3840), so for now, I set points C and D at the corners of the image.

I wanted points A and B to be exactly on the posts at the roadside. I adjusted their positions with a few attempts.

Now I move points C and D outward so that the lines align with the edge of the road. It takes a few more attempts to get it just right.

I know it's a rather monotonous and manual method, but it works.