Utilizing super resolution to increase disparity resolution and better stereo depth estimation for fisheye images

Abstract

This work attempts to improve the stereo fisheye depth estimation for far away objects. Fisheye cameras have a great advantage over traditional cameras in that they have a wider field of view (FoV) to perceive its surroundings. However, due to the larger sampling rate from the wider FoV, the wide FoV information is compressed into a limited sensor array. When performing stereo depth estimation using fisheye cameras, the depth is estimated using disparity between corresponding pixels. When attempting to estimate depth for far away objects, low resolution images lack depth resolution as the disparity resolution is reduced

this effect is more apparent in fisheye cameras as these cameras have lower spatial resolution. To improve the depth resolution of far away objects in fisheye camera stereo depth estimation, the proposed research applies super-resolution on fisheye images. By performing super resolution, the disparity resolution is increased as well as the depth resolution.