Underwater Multi-sensor Calibration,Reconstruction And Application

Underwater sensing technology plays a vital role in human’s continuous exploration of underwater production and scientific research activities,and is pivotal in underwater mineral exploration,underwater terrain mapping and underwater rescue.However,the fluctuation,turbidity and absorption of light in the underwater environment,coupled with the refraction of light under multi-media conditions,all restrict the various sensors placed in the sealed cabin to obtain accurate measurement parameters.In addition,using a single sensor to complete underwater scanning and positioning tasks faces challenges such as lack of information and accuracy to be improved.In order to solve the problem of calibration and positioning of underwater sensors,this paper starts from the underwater refraction camera calibration,and introduces sensors such as line laser and IMU to jointly carry out underwater reconstruction and positioning after obtaining accurate camera parameters.The main work of this paper is as follows:(1)Given that the limited field of view of the camera in the sealed cabin in the underwater environment,and the refraction of light under multi-media conditions affects the imaging accuracy,this paper proposes an underwater camera refraction calibration algorithm using a spherical waterproof cover.In the process of converting the hemispherical surface equation from the dome reference frame to the camera reference frame,the compensation of translation and rotation is added,and then the optical path expressions in each medium are derived,and finally the center offset is optimized by using the three-dimensional point orthographic projection model and the parameters of the two-dimensional point and the model are obtained by optimizing the back-projection iterative method,so as to calculate the re-projection error and improve the calibration accuracy of the underwater dome port camera.(2)Regarding the problems of low precision and pathological correspondence in monocular vision reconstruction,this paper uses camera-line laser to reconstruct underwater objects in 3D.After obtaining accurate underwater camera parameters,the projected laser stripes are processed and extracted first,then the corresponding laser plane is fitted,and finally the intersection point of the direction vector of the light in the water and the laser plane is obtained as the surface contour point of the object.The complete reconstruction result can be obtained by splicing the results of each scan to the first frame by the way of panning and scanning.(3)Aiming at the problem that the feature extraction is difficult to meet the tracking requirements under the condition of weak underwater texture and insufficient illumination,this paper designs an underwater visual-inertial odometry combined with point-line features.The front-end extracts ORB features and uses optical flow to track each key frame,and at the same time extracts LSD line features for matching;the back-end uses a sliding window-based method to optimize IMU measurement residuals and point-line feature reprojection residuals.Experimental results on official datasets and data collected underwater demonstrate the reliability of the proposed algorithm.