Simultaneous Localization and Mapping Techniques with AI-Based Stereo Camera

Abstract

This thesis involved incorporating two distinct stereo camera systems, namely ZED-X and ZED-2i, through the utilization of SLAM technology to create a three-dimensional representation of an environment. The primary goal was to enable the systems to generate a 3D map using point clouds and accurately track the position, specifically for robotics applications. Two experimental setups were created with distinct configurations to facilitate comparison and interpretation of the results. The utilization of the SLAM technology within the robotic operating software played a pivotal role in augmenting the system's performance. The system was structured by developing dataflow diagrams and establishing a workspace in ROS. Both systems underwent testing in an indoor setting, while the ZED-2i system was additionally tested outdoors. The experimental findings for this thesis were remarkable, as the ZED-X demonstrated precise visual outcomes. Within the ZED-2i system, there were instances of erroneous identifications and a reduced level of point cloud density compared to the ZED-X. During the testing phase, both setups exhibited favorable and elevated rates of detecting the depth and positioning of objects. Despite the limitations observed in both setups during the experiments, these challenges can be addressed through additional investigation and more comprehensive testing. One of the proposed recommendations for future research was the implementation of the ZED-2i and ZED-X stereo cameras for a robotic system.