Robustness to lighting variations: An RGB-D indoor visual odometry using line segments

Abstract

Abstract-Large lighting variation challenges all visual odometry methods, even with RGB-D cameras. Here we propose a line segment-based RGB-D indoor odometry algorithm robust to lighting variation. We know line segments are abundant indoors and less sensitive to lighting change than point features. However, depth data are often noisy, corrupted or even missing for line segments which are often found on object boundaries where significant depth discontinuities occur. Our algorithm samples depth data along line segments, and uses a random sample consensus approach to identify correct depth and estimate 3D line segments. We analyze 3D line segment uncertainties and estimate camera motion by minimizing the Mahalanobis distance. In experiments we compare our method with two state-of-the-art methods including a keypoint-based approach and a dense visual odometry algorithm, under both constant and varying lighting. Our method demonstrates superior robustness to lighting change by outperforming the competing methods on 6 out of 8 long indoor sequences under varying lighting. Meanwhile our method also achieves improved accuracy even under constant lighting when tested using public data.

RGB-D SLAM with Structural Regularities

“…Differently from the TUM RGB-D and ICL NUIM datasets, the TAMU dataset [36] provides large-scale sequences (constant lighting), covering a whole floor or multiple rooms. While it does not provide ground-truth camera poses, the start and end point are the same, which can be used to evaluate the overall drift by computing the final position errors.…”

A Comparative Study for Indoor Planar Surface Segmentation via 3D Laser Point Cloud Data

“…Fischler and Bolles (1981) introduced the well-known model-based RANSAC algorithm used in segmenting planar surfaces in point cloud data. Since the RANSAC does not need to know the relation between neighbor points, it could be applied to both organized point clouds acquired by RGB-D camera (Lu and Song, 2015) and unorganized point cloud data captured with 3D LiDAR or laser sensors (Xu et al, 2016) without any preprocessing step. Moreover, Schnabel et al (2007) presented a RANSAC-based method for automatically segmenting fundamental shapes such as plane, cylinder, and cones in both mesh and point cloud data.…”

Semi-dense visual odometry for RGB-D cameras using approximate nearest neighbour fields

“…Line, curve based and semi-dense methods: Lines are alternative features to points and have been widely used in many VO and SLAM frameworks such as [13], [14]. One reason is that lines are abundant in man-made structures and environments, and do not depend on sufficient texture.…”

“…In order to show that our method is capable to work in relatively large-scale environments, we provide reconstruction results on two sequences from the TAMU RGB-D datasets [14] and ICL-NUIM synthetic datasets [27]. As shown in Fig.…”