Unmanned Ship Ranging and Error Correction Method Based on Monocular Vision

Aiming at the problem that traditional unmanned ship mainly uses lidar, ultrasound and other devices to detect water surface obstacles, which cannot accurately identify the specific types of obstacles, and accompanied by high cost, low accuracy and large impact by sea conditions, a monocular vision unmanned ship ranging and error correction method is proposed. First, the obstacle is identified and segmented by the trained Mask R-CNN model, the depth information of the obstacle is obtained by using the monocular ranging algorithm based on ground constraint, obtaining the posture of a ship using a three-dimensional electronic compass and correcting the errors. In the actual water scene, the average relative error of ship ranging is 4.64%, and the absolute error is controlled with in 2m in the range of 0-30m. Ranging validation on CoppeliaSim simulation platform shows that the average relative error is 9.30% and the average absolute error is 0.89 meters in the range of 0-30 meters. The monocular ranging algorithm with error correction has higher ranging accuracy and can meet the obstacle avoidance needs of small unmanned ship at low speed.