Optimize feature detection and tracking

asked 2019-10-17 16:39:07 -0500

Hi all,

I'm newly getting into OpenCV and don't quite know how to ask that question. I'm analyzing a video stream to find a picture inside the video frame and then I track the movement of the picture inside the video frame. I'm talking about movement in terms of 3D coordinates. I would like to overlay the tracked picture with a video that always covers/overlays the picture.

In theory that is working alright but somewhat slow. Detection is slower than tracking but tracking is still slow.

Now to the question. How can I get the detection and tracking faster? I'm thinking if it might be possible to use the dimensions of the input image and the relation of the feature points to each other to somehow crop the area that I would need to scan for features during detection. But that is just a crazy idea I have. I'm open to suggestions. The original code for the tracker is here: https://github.com/artoolkitx/artoolk...

Any hint highly appreciated and may thanks in advance.

Kind regards

edit retag flag offensive close merge delete

Comments

Use a native object detector with cuda. They can be pretty fast on decent hardware.

holger gravatar imageholger ( 2019-10-18 02:22:52 -0500 )edit

Thanks @holger. Unfortunately, that won't work for me as I'm aiming for mobile and small hardware like Raspberry PI.

Thor_Bux gravatar imageThor_Bux ( 2019-10-20 19:24:26 -0500 )edit