Object tracking

[eberlawrence]

Hello everyone,

I learned about HTM a while ago. I am still studying and trying to catch up with the theory
and its implementation.

I’ve been thinking if it is possible to apply HTM to the problem of object tracking.
I would like to identify an object and then create predictions of its location in space
given the observations of how it is moving. This could be useful in cases where the object
becomes ocluded for a moment and returns to view after a while or for controlling a robotic
arm to move and block a ball moving towards it.

From your experience, do you believe it is possible to use HTM for such application?
Is it a good idea to use HTM for this?
Would you suggest any source materials or have some ideas on how to implement this object tracking?

Looking forward to hear your thoughts.
Thank you!

[rhyolight]

Definitely see https://github.com/numenta/nupic.geospatial. Although it’s an old demo.

[rhyolight]

@eberlawrence Did you watch the video? Is this something you think you might try? There is an example codebase and everything you could use.

[marty1885]

Maybe my previous project can help? A GridCell Decoder will be needed tho; to convert HTM predictions back to location coordinates. I’m more than happy to build one if someone needs it.

Path anomoly detection using Grid Cells and Temporal Memory

I’m surprised that no one has ever tried grid cells as location encoders (like to the old GeoSpatial Encoder) for anomaly detection. So it became an experiment I ran for the final project for my ML course. Here is a screen shot for your viewing pleasure. And letting me to explain things more clearly. [image] On the left you see a white and red circle revolving in a circle. That is the target. HTM will try to learn lt’s path and generate anomaly signals when even the path is altered. On the …