SLAM Using 3D Reconstruction via a Visual RGB and RGB-D Sensory Input

Abstract

This paper investigates simultaneous localization and mapping (SLAM) problem by exploiting the Microsoft Kinect™ sensor array and an autonomous mobile robot capable of self-localization. The combination of them covers the major features of SLAM including mapping, sensing, locating, and modeling. The Kinect™ sensor array provides a dual camera output of RGB, using a CMOS camera, and RGB-D, using a depth camera. The sensors will be mounted on the KCLBOT, an autonomous nonholonomic two wheel maneuverable mobile robot. The mobile robot platform has the ability to self-localize and preform navigation maneuvers to traverse to set target points using intelligent processes. The target point for this operation is a fixed coordinate position, which will be the goal for the mobile robot to reach, taking into consideration the obstacles in the environment which will be represented in a 3D spatial model. Extracting the images from the sensor after a calibration routine, a 3D reconstruction of the traversable environment is produced for the mobile robot to navigate. Using the constructed 3D model the autonomous mobile robot follows a polynomial-based nonholonomic trajectory with obstacle avoidance. The experimental results demonstrate the cost effectiveness of this off the shelf sensor array. The results show the effectiveness to produce a 3D reconstruction of an environment and the feasibility of using the Microsoft Kinect™ sensor for mapping, sensing, locating, and modeling, that enables the implementation of SLAM on this type of platform.