静态物体的视频稳定

Question

Hello to all the engineers!

I have the grayscale image of my static object which I want to have fixed place on the image during the acquisition process. The brightness constancy is not guaranteed for the image object. Here is sample image attached.

Moreover, I desire to get stabilization when camera shifts the image for 20-30% of its width. The algorithm should be also fast enough to be implemented in embedded computer for real-time acquisition either in CPU or DSP.

What algorithm or algorithm chain should be used for this purpose? What woudld you recomment to read? I thought to use the reference frame and finding the optical flow, then making inverse affine transform.

I tried to use OpenCV optical flow (pyramid Lukas-Kanade) with some special parameters, the result was satisfactory when the camera movements did not excee 10% pixelx of image in each dimension.

Answer 1

Looks like your example image has some well-defined edges, which you could find regardless of brightness. Edge finding isn't particularly expensive, and from there it's a simple comparison.

I don't know what you mean by "camera shifts the image by 20%". Common operations for camera's include pan, tilt and zoom. (aka "PTZ")