How to determine the source coordinates better in computer vision speed estimation

[dearMOMO]

Search before asking

• I have searched the Supervision issues and found no similar feature requests.

Question

blog https://blog.roboflow.com/estimate-speed-computer-vision/

Is there A good tool to calculate the coordinates of A, B, C, D? Thank you very much

Additional

No response

[likith1908]

@dearMOMO did you find any solution to this?

[aliyilma]

Good question, let me diversify the topic a bit according to my own work. Is there an easy method for perspective correction/determination? In my own work, sometimes I need to do perspective correction of the video, sometimes I just need to transform for object traces. It can be done somehow with OpenCV but it complicates the code and I don't get very good results. It would be nice if there was an easy method in the Supervision library. For example, detection/trajectories could be transformed according to the rectangle drawn on the screen and given its dimensions, and even projected in a second image window (this could be useful for other annotation methods like tracks).

There is a very nice automated implementation in the case study, especially for dynamic camera angles.
https://www.youtube.com/watch?v=yJWAtr3kvPU
The tracker software also has an interactive perspective conversion tool for video and coordinates
https://physlets.org/tracker/help/video_filters.html

[SkalskiP]

Hi 👋🏻 I used keypoint detection models to do it.

football-ai.mp4

[aliyilma]

hi, awesome work! looked at your code () and couldn't find how to project players to the 2D field, or it might be because my coding skills not very good. in my case, i am working on different crowded scenes and keypoint detection might not be useful with those.

[SkalskiP]

@aliyilma this is my old demo project; the code for this project will be hosted here soon.