Image Based River Navigation System of Catamaran USV with Image Semantic Segmentation

Abstract

USV’s river navigation system is very complex, including USV’s environmental perception, USV’s positioning, obstacle avoidance, and path planning. Different sensors are required in various applications, such as lidar or cameras in environmental perception, GPS or IMU in positioning, and ultrasonic sensors or cameras in obstacle avoidance. As an intuitive and affordable sensor, the cameras can be used for environment perception, obstacle recognition, and distance estimation for component stereo vision systems over a catamaran USV. This paper proposes an image based USV river navigation system. The main functions include riverbank line recognition and distance estimation. Aiming at the inaccurate of traditional riverbank line detection, we design a line segment detection algorithm for riverbank based on image semantic segmentation, which can effectively reduce the interference of the object’s texture on edge recognition while ensuring the object edge recognition. Subsequently, according to the built stereo vision system, the distance of the detected riverbank line is estimated for following navigation, obstacle avoidance, or other operations. We further demonstrate the designed catamaran USV and its stereo vision system. Compared with the traditional unsupervised straight line detection algorithm, its riverbank line detection function has been significantly improved, and its real-time ranging process is also more intuitive.