Visual-inertial curve SLAM

Abstract

In this dissertation, we present a simultaneous localization and mapping (SLAM) algorithm that uses B\'{e}zier curves as static landmark primitives rather than feature points. Our approach allows us to estimate the full 6-DOF pose of a robot while providing a sparse structured map which can be used to assist a robot in motion planning and control. We demonstrate how to reconstruct the 3-D location of curve landmarks from a stereo pair and how to compare the 3-D shape of curve landmarks between chronologically sequential stereo frames to solve the data association problem. We also present a method to combine curve landmarks for mapping purposes, resulting in a map with a continuous set of curves that contain fewer landmark states than conventional point-based SLAM algorithms. We demonstrate our algorithm's effectiveness with numerous experiments, including comparisons to existing state-of-the-art SLAM algorithms.

A notable contribution of this dissertation is to apply our SLAM algorithm to a river setting to localize a canoe and create a sparse structured map of the border of a river. To accomplish this task, the dissertation presents a novel vision-based algorithm that identifies the boundary separating water from land in a river environment containing specular reflections. Our approach relies on the law of reflection. Assuming the surface of water behaves like a horizontal mirror, the border separating land from water corresponds to the border separating 3-D data which are either above or below the surface of water. We detect a river by identifying this border in a stereo camera. We start by demonstrating how to robustly estimate the normal and height of the water's surface with respect to a stereo camera. Then, we segment water from land by identifying the boundary separating dense 3-D stereo data which are either above or below the water's surface. With the border of water identified, we validate the proposed river boundary detection algorithm by applying it to a chronologically sequential video sequence obtained from the visual-inertial canoe dataset. Additionally, we use our SLAM algorithm to create a sparse structured map of the shoreline of a river.