Image-guided Landmark-based Localization and Mapping with LiDAR

Abstract

Mobile robots must be able to determine their position to operate effectively in diverse environments. The presented work proposes a system that integrates LiDAR and camera sensors and utilizes the YOLO object detection model to identify objects in the robot's surroundings. The system, developed in ROS, groups detected objects into triangles, utilizing them as landmarks to determine the robot's position. A triangulation algorithm is employed to obtain the robot's position, which generates a set of nonlinear equations that are solved using the Levenberg-Marquardt algorithm. The presented work comprehensively discusses the proposed system's study, design, and implementation. The investigation begins with an overview of current SLAM techniques. Next, the system design considers the requirements for localization and mapping tasks and an analysis comparing the proposed approach to the contemporary SLAM methods. Finally, we evaluate the system's effectiveness and accuracy through experimentation in the Gazebo simulation environment, which allows for controlling various disturbances that a real scenario can introduce.