A sky region segmentation method for outdoor visual-inertial SLAM

Abstract

Visual Simultaneous Localization and Mapping (VSLAM) can build a map of an unknown environment and perform self-localization simultaneously by computer vision, which has been playing an important role in UAV, UGV and mobile robot etc. At present, slow calculation speed and low accuracy are still the main problems in outdoor scene. In this paper, we segment the image of the visual part of the sky region. At first, threshold segmentation is carried out for gradient images, then boundary line extraction is carried out for segmented images using improved energy function, and then refined region stripping is performed using polynomial fitting. This algorithm avoids feature extraction of large sky region and further optimizes the accuracy and computing speed of visual inertial odometer in outdoor scene. Experiments show that the new system has improvements in translation and rotation errors.