B-Splines Based Data Association Method for the SLAM Problem in Complex Environments

Abstract

This paper presents a new proposal to solve the problem of data association for SLAM used to build of maps of complex environments. The main idea of this proposal is to use B-spline curves as a way to describe obstacles with complex geometries found in the environment and use the information contained in them to find characteristic points that may be associated. The use of this information for a more accurate association process is one of the major contributions of this work, because a robust association has a direct impact on the localization of the robot and thus the quality of the final map. The data association problem was initially addressed by comparing the control points that form both the curves representing the obstacles observed at a given time, and those that represent the obstacles stored in the map being constructed, relating those that are close enough. Then, the curvature of the related B-spline is obtained to extract characteristic points (inflection points and corners) contained in the curves. Finally, the matching information will be used to correct the position of the robot and the detected obstacles. We carried out numerous experiments by using real and simulated information in order to validate the processes and algorithms proposed in our approach. Our method achieves a great precision in map construction of complex environments, which is nearly impossible with techniques that currently exist.