LiDAR : camera fusion for SLAM of autonomous robots or drones in indoor environments

This master’s thesis is part of a long-term academic project, launched by the Polytechnic School of Louvain, with the aim to develop technologies that would enable low-cost UAVs to navigate autonomously in indoor environments. In the previous years, several master theses tried to bring a robust solution for localisation and mapping of autonomous drones based on the AR.Drone 2.0 from Parrot, equipped with a set of sensors including a monocular camera, an inertial measurement unit and various localised range finders. Our master’s thesis differs from previous these and work on homemade modular and portable platform to develop a sensor system for the SLAM without the constraints from the equipment of the (outdated) AR.Drone 2.0. In this master's thesis, we will the study the potential of a system combining a monocular camera with a LiDAR sensor. The main goal is to map out the overall possibilities and potential implementation with the addition of a LiDAR to the sensor system. After a review of the State-of-the-art regarding SLAM and relevant sensors, we propose a first implementation of a sensor system based on monocular camera with a 2D scanning LiDAR sensor. This system combines the depth information from the LiDAR and the visual information from the camera to create 3D point clouds and depth images which can be fed to state-of-the-art visual SLAM algorithms. The sensor system shows interesting results and is first step toward a LiDAR based SLAM method. The solution works but not perfectly. One of the main issue is the computational time to establish the depth images. Therefore, we will also address the shortcomings and discuss interesting directions going forward.