Real Time Image Stabilization

Question

I'm interested in implementing real time image stabilization and was wondering if anyone had any suggestions for how to achieve such a task. I have a camera and a Bosch BNO055(gryo) hooked up to an embedded processing board, and was looking at possible techniques such as: http://hci.stanford.edu/cstr/reports/2011-03.pdf

To be honest this method is a bit more complex than I was originally hoping for and i'm not to interested in correcting rolling shutter.

I'm interested if anyone had any other ideas, or had come across any other papers that they would be willing to point me at. Code examples would be awesome.

Thanks

Answer 1

The OpenCV provides a some usefull implementations such as videostab. A very simple and fast algorithm would be:

1. Estimate feature correspondences from consequtive frame. ( common methods apply are SIFT, SURF, FAST, if these are to slow you can estimate motion vectors from a regular samples grid done by the pyramidal Lucas Kanade method)
2. Estimate a homography based global motion model e.g. with RANSAC (OpenCV: getHomography function)
3. Use such a homography to warp your image (OpenCV: warpPrespective function)