Development of binocular vision system for identification

Abstract

Unmanned Surface Vessels (USVs) are transforming maritime operations. These self-navigating systems rely on advanced sensors, including GPS, LiDAR, and cameras, to independently traverse and make decisions on water without human intervention. Their applications span diverse fields, from military surveillance to environmental research and oceanography. The Maritime RobotX Challenge, a prestigious global competition held biennially, serves as a platform for advancing autonomous maritime technologies. This competition attracts university teams worldwide, fostering collaboration between academic institutions and the maritime industry. Participants develop cutting-edge ASVs to tackle complex tasks, pushing the boundaries of innovation in autonomous marine robotics. In the context of this competition, our research centers on designing a binocular vision system specifically tailored to address RobotX challenges, including object recognition and localization. The proposed system introduces a novel methodology that combines a top-down image sampling strategy, hybrid feature extraction methods, and the Restricted Coulomb Energy (RCE) neural network for effective cognitive processing and recognition. By leveraging stereo vision techniques, the system calculates accurate three-dimensional coordinates of detected objects, enhancing spatial awareness. To complement the RCE neural network, YOLOv8 was employed for real-time object detection and classification. Its efficient architecture enables precise identification of key maritime objects, such as buoys and light towers, which form the foundation for subsequent localization tasks. Additionally, a color sequence recognition module was incorporated to detect specific patterns displayed on navigational aids like light towers. This process ensures robust identification of color-coded sequences by focusing on designated regions, effectively mitigating background interference. This integrated system showcases the synergy of cutting-edge neural network algorithms and stereo vision technology, delivering a reliable and adaptive solution for autonomous maritime applications within the RobotX Challenge framework.