A study of 2D features for 3D visual SLAM

Abstract

The use of 2D features in computer vision has had a great impact in lots of applications. For example, the combination of those features together with 3D data has helped to solve the Simultaneous Location And Mapping (SLAM) problem in real time. Nowadays, there are several interesting feature detectors and descriptors with different characteristics: processing time, robustness against lightning conditions, changes in point of view, scale, etc., and every day it appears more and more. In this paper, a deeper study about several of these detectors and descriptors is done. Several interesting graphs where we can separate distances with respect to axis and angles have been analysed. This study helps to make decisions about which are better for a given application. The study has been done with a low cost sensor as Kinect installed on robotics arm to control the movements with accuracy. Furthermore, finally, real scenario reconstruction is shown using Kinect camera and the visual features analysed in the study.