An Extensive Approach to Features Detection and Description for 2-D Range Data Using Active B-splines

Robust LIDAR-Based Localization in Underground Mines

“…The falko method includes only simple computations and is therefore well suited for real-time applications. This is experimentally verified in [110], where computation times for flirt and falko are compared to their own contribution bid. falko is roughly 50 times faster than flirt and 100 times faster than bid.…”

“…The computer vision community has successfully built an arsenal of feature extraction methods for the purpose of getting robust data associations in noisy data with low computational effort [56]. Procedures developed for image and point cloud processing have also been adopted for feature extraction methods specifically adapted to 2D lidar data [50,107,110]. These 2D lidar feature extraction methods are proven to perform well in clean indoor environments.…”

“…The B-spline based interest point detector (bid) is introduced in [110]. It uses the same concept as falko by extracting high curvature candidate points using a set of neighbor points defined as in (2.46) with a radius growing with measured range as in (2.47).…”

“…As for detectors, there has been substantial work performed in the computer vision community [16,56,61], and in [63] the sift descriptor is used for 2D range data. However, the results in [50,110] indicate that stable detection of feature points is much more important than a sophisticated description to obtain good feature matching in 2D laser data. Therefore the main focus of the reminder of this section is on the descriptors accompanying the flirt, falko, and bid detectors in [107], [50], and [110], respectively, all originating from the same concept of a shape context.…”

“…This is suitable for their place recognition purpose, but it cannot really be viewed as a local keypoint feature descriptor. Nevertheless, one of the best performing descriptors in the experiments conducted in [15] is based on the concept of shape contexts, which is also the case for the descriptors used in state-of-the-art range data feature extraction work [50,107,110]. The concept of shape contexts is developed for object recognition in images in [8].…”