Motion compensated automatic image compositing for GoPro videos

Abstract

Image composition for GoPro videos captured in the presence of significant camera motion is a manual and time consuming process. Existing techniques typically fail to automate this process due to the wide-capture field of view and high camera motion of such videos. The proposed method seeks to solve these problems by developing an image registration algorithm for fisheye images without expensive pixel warping or loss of field of view. Background subtraction is performed to extract moving foreground objects, which are noise corrected and then layered on a reference image to build the final composite. The results show marked improvements in accuracy and efficiency for automating image composition.