Robust estimation of camera motion parameters

摘要

In this paper, we propose an affine parameter estimation algorithm from the block motion vectors for extracting accurate motion information with the assumption that the undergoing motion can be characterized by an affine model. In this method the spurious motion vectors are discriminated from the genuine ones using specific error detection criteria. The proposed method has two notable features: firstly, the motion vectors can be estimated accurately without error, and secondly, the blocks used in the estimation are in close proximity to each other and are all selected from those parts of the video frames that are of equal medium depth. This is an important requirement for estimation of the various parameters of camera movements in video stabilization. Experimental results show that our approach is very effective in extracting the model parameters.