A Novel Visual Measurement Method for Three-Dimensional Trajectory of Underwater Moving Objects Based on Deep Learning

For deep sea equipment suspension installation used in marine engineering, the multi-camera video motion analysis method is used to calculate the three-dimensional underwater trajectory of the underwater engineering structure. Considering difficulty in underwater modeling caused by the problems of light scattering and refraction under water, the camera imaging model on land is no longer applicable in water, and a new underwater camera imaging model needs to be proposed. This paper introduces an underwater camera imaging model with light refraction, studies the calibration method of the internal and external parameters of the underwater camera, and improves the multiscale rotation dense feature pyramid convolutional neural network to detect the position of the target object in the image. The underwater motion videos of the target produced by three fixed underwater cameras are optimized to fuse and calculate the trajectory of the underwater target. This method is suitable for large-scale motion of underwater objects and can obtain more accurate trajectories. Experimental analysis and data comparison have verified the effectiveness of the method.