I’m scoping out a project that requires a few cameras tracking people playing a ball game on a small court. We need to follow the ball and have some latitude to zoom in. From the footage, we want to automate the creation of highlight reels, add graphics and pull statistics. I’d also like to integrate Gen AI such as Stream Diffusion to add a layer of animation on top of the footage.
There are a few products out there that do just this: Spiideo and Veo for example. But they appear to be closed off black boxes. You typically subscribe to their services and access the video and data through their software. This makes me worry that we wouldn’t be able to plug the material into our own systems.
So I’m wondering first if there are any open source versions, or if anyone’s had success hacking into these?
Or, (shudder) what about developing our own? How advanced are these commercial versions compared to the machine learning tracking we have available now?
I’d love to hear anything you’ve learned along the way.