Development of a binocular vision system for guidance

• Main Author: Song, Jieni
• Other Authors: Xie Ming
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2025
• Subjects: Engineering; Unmanned surface vessels; Stereo vision; YOLOv8; 3D object localization; RCE neural network
• Online Access: https://hdl.handle.net/10356/182400
• Institution: Nanyang Technological University

Unmanned Surface Vessels (USVs) are increasingly deployed in modern maritime operations, leveraging advanced technologies such as stereo vision and neural networks to achieve autonomous navigation and robust environmental interaction. This study focuses on the development of a binocular vision system integrated with machine learning algorithms to enhance USV perception and localization capabilities. The research is framed within the context of the Maritime RobotX Challenge, an international platform that encourages innovation in autonomous maritime systems. This dissertation evaluates two recognition methodologies — Restricted Coulomb Energy (RCE) neural network and YOLOv8—for object detection and localization. While both approaches are analyzed for their respective strengths and limitations, one is ultimately implemented for real-world application. The system employs the ZED 2i binocular camera to achieve precise three-dimensional object localization, leveraging stereo vision to capture depth information and improve spatial awareness in dynamic maritime environments. Extensive experimental evaluations validate the effectiveness of the system in detecting and localizing maritime objects such as buoys and light towers, with results demonstrating strong precision and recall rates. The research also explores hardware integration and system optimization to ensure robust performance in real-world scenarios. By combining high-resolution stereo imaging with advanced machine learning algorithms, the system effectively detects, classifies, and localizes objects in complex and dynamic environments. The findings contribute significantly to advancing the design and deployment of USVs, enabling them to perform critical tasks such as navigation, obstacle avoidance, and environmental monitoring. Future work could explore integrating additional sensors like LiDAR or sonar and optimizing algorithms to further enhance system reliability, versatility, and overall performance in diverse maritime scenarios.