In a current project (which might see commercial use in the futere) I need to track people and objects in camera images. A little bit of research brought up the "Discriminative Scale Space Tracker" (DSST), which is detailed in the paper: Accurate Scale Estimation for Robust Visual Tracking. This algorithm is based on the MOSSE Tracker (Visual Object Tracking using Adaptive Correlation Filters)

Before implementing everything myself, I looked around and found an open source implementation. This page told me:

The code using linear correlation filters may be affected by a US patent. If you want to use this code commercially in the US please refer to http://www.cs.colostate.edu/~vision/ocof_toolset_2012/index.php for possible patent claims.

I tried to read the patent documents: http://www.google.com/patents/US20100322534?hl=en and http://www.google.com/patents/US8520956?hl=en. However, now I am really confused and don't really know what I am allowed to do. The authors of the github page also seemed unsure if the code is affected by the patent.

Does the patent concern the overall idea of using correlation based filtering? Some of the claims seem very basic to me, e.g.:

dividing each video frame into a plurality of detection regions and applying the final filter to each region.

Which is essentially what every object detector does.

The final question would be: Am I allowed to use DSST in the project either in the already open source implementation or in a new implementation solely based on the DSST-Paper? Or is the method too closely related to the patented methods?

Thank you!

1 Answer 1


It is impossible to give a complete answer without carefully considering what the patent claims are and what your new code or product does. Since people here don't know about your project, they can't conduct the necessary analysis.


dividing each video frame into a plurality of detection regions and applying the final filter to each region.

is only part of a claim. It is not a claim by itself.

Each claim, even if it's hard to read, will be essentially a list of elements or features.

Does every single thing (every feature or limitation) in a claim appear in your code or product?

If "yes", then you may have a problem. This analysis should be repeated with each claim.

If there is something in a claim that you're not doing, then you're OK.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .