It really depends on the ratio of the ball size to the frame size. My ball tracking algorithm works surprisingly well that it can detect a soccer ball near the far corners which I think is fairly similar to a tennis ball in a close-up.
I only tested it on cylindrical videos but I think it should work on normal videos too. It is also possible to transform the cropped video into the perspective view but I haven’t done that.
This looks amazing! Currently doing this by reframing the final video using a lot of manual keyframing. Any idea how long until we can try this out, even as an alpha/beta?
Anything we can do to help? I can provide example videos including daytime/nighttime games if it helps.
Yes I would like to see how well my AI tracking implementation work on your videos. Please post them here or PM me the links if you don’t want to share publicly.
I’m currently using another program (Autocam) to take the stitched videos and then crop them to 1080 with ball tracking It does a decent job, but often times it will focus on total movement vs. ball movement I feel like. So it works well for flag football, but makes more than a few mistakes in soccer on a full-size field. Would love to compare the two results, especially because of how well the stitching works and all the support you provide.
I’m uploading another one right now too. Should be 2 daytime and 2 nighttime
These are all recorded using the gamsod at 4k60 and the 3D printed mount
In all of them one side is a bit blurrier than the other, pretty sure one of the cameras I received was defective, so I’m working on a replacement. Hopefully those are still clear enough for the tracking to work.