gnebehay/OpenTLD

Is there a way to track multiple/all moving objects?

Closed this issue · 3 comments

The API in the README looks like you have to specify an initial bounding box of an object you want to track. Is there a way to track more than one of these objects?

Unfortunately not. The best that you can do at the moment is to start multiple instances with different initialisation targets. There also no plans extending to incorporate any form of multi-target tracking in the near future.

[someone who knows the code should confirm/deny this, I'm remembering the parts of the paper]

It's worth noting that the Learning part in the TLD algorithm relies on the concept that there is 1 positive sample at any given time - and uses the rest of the image as negative samples.

So, if you have several very similar objects that you want to track in the video, you could have real problems.

@racamirko That is correct, thanks for pointing that out.