Parallax Mitigation for Real-Time Close Field Video Stitching
Alex J Watras, Jianwei Ke, Zhanpeng Zeng, Jae-Jun Kim, Hewei Liu, Hongrui Jiang, Yu Hen Hu · 2017
In this paper, we consider the task of stitching video frames captured by an array of micro-cameras in order to create an enlarged field of view video for surgical applications. We propose a method for mitigating the problem of parallax when dealing with non-planar scenes with moving objects that cross boundaries of adjacent views, causing discontinuities at view boundaries of the stitched video. The discontinuities created by parallax are especially noticeable when the object has long straight lines that cross the boundary. In this work, we propose a method of aligning the edges of objects affected by the parallax in the scene using epipolar geometry and then applying secondary transformations to correct parallax discontinuities. Our method leverages the regular shape of the object and is suitable for surgical applications where few if any feature points may be extracted from the appearance of the objects.