Satellite Navigation using FishEye Sky Segmentation and Deep Learning

A novel navigation satellite positioning method based on deep learning and fisheye camera image segmentation is proposed in this paper, aiming to address the issue of satellite signals that can easily be affected by various environmental factors such as object occlusion in traditional satellite navigation and positioning methods in complex outdoor scenes. It uses a fisheye camera facing the sky to assist in determining whether the received satellite signal is a non-line-of-sight (NLOS) signal or not. After segmenting the sky and non-sky regions of the fisheye camera image, the satellite position is projected onto the pixel plane of the fisheye image, and the visibility of the satellite is determined based on the projection results, which can remove the NLOS signals, and ultimately achieve correct positioning results. Compared to traditional satellite navigation positioning methods, the method proposed in this study can improve positioning accuracy by an average of about 31.9%, with good results.