A Fast Underwater Calibration Method Based On Vanishing Point Optimization Of Two Orthogonal Parallel Lines

Underwater optical imaging,underwater stereo matching and three-dimensional(3D)reconstruction of the underwater environment are all based on the underwater vision system.Calibration technology is the most critical and complex part of underwater high-precision three-dimensional measurement.The traditional calibration method requires at least three images from different viewing angles,and there are problems such as too many calibration constraints and too long calibration running time.At present,the direct calibration method based on the vanishing point in the air only needs one image.By extracting the geometric features in the image,the calibration can be completed according to the geometric constraints between the vanishing point and the calibration matrix.Since this method only needs to collect one image,it has high requirements on the shooting quality of a single image,and it is necessary to pre-process the underwater image to remove the influence of water ripples before calibration;secondly,the direct calibration method based on the vanishing point is used in the air without considering the underwater imaging model,resulting in too large error in the calibration parameter results,which is not suitable for underwater application scenarios.Aiming at the problem that the single image shooting quality required by the vanishing point calibration method is high,this paper proposes a water ripple elimination method based on the fusion of ripple characteristics and environmental characteristics,and performs enhancement preprocessing on the collected single image.Firstly,the image is subjected to secondary decomposition based on sym2 wavelet,and the image frequency bands based on ripple and environmental characteristics are separated;secondly,the image frequency bands based on ripple characteristics are detected by Markov random field,and then the water ripples are removed based on the expectation maximization method.Then,the water ripple area is tested and judged based on the equation maximization,and the above detection and removal methods are iterated until the elimination threshold is met;for the image frequency band based on environmental characteristics,the exposure based on the underwater imaging model is used to estimate the exposure,and then the bilateral filtering method is used to estimate the exposure.Perform denoising,then evaluate exposure based on Gaussian model,then perform band enhancement on grayscale linear transformation,and finally perform inverse wavelet transform on the enhanced ripple and environmental image bands.This method reduces the effect of artifacts caused by water ripples,improves the quality of high-frequency segmentation of underwater images,and can display more complete underwater details.The image mean square error value decreased by 18.04%,and the image quality measurement value based on contrast change increased by 17.03%.The experimental results show that this method enhances the contrast and brightness of low-quality water ripple images,and solves the problem of less detailed feature information in underwater images under the influence of water ripples.Aiming at the problem of image distortion caused by the refraction of light in the three media of air,glass and water,which will affect the accuracy and speed of the calibration of the underwater monocular camera,the traditional calibration technology needs to introduce high-order distortion coefficients and multiple pose pictures,which leads to the calibration time.too long.In the end,this paper proposes a fast underwater calibration method based on two sets of mutually orthogonal parallel lines corresponding to the corresponding vanishing point optimization.Firstly,the distortion correction based on the single-layer photorefractive model of glass-water is carried out on the single underwater image;secondly,the Hough transform is used to extract mutually perpendicular parallel lines,and the initial image coordinates corresponding to the vanishing point are solved,and then two groups of two groups are constructed based on distance optimization.The precise vanishing point corresponding to the mutually orthogonal parallel lines is used to obtain the underwater focal length of the camera based on the above vanishing point.Finally,the equivalent image of the underwater image in the air is obtained based on the air-underwater equivalent focal length model,and then the external image of the camera is obtained.Parameters(rotation matrix and transformation matrix).This method only needs to collect one image;the algorithm only needs two sets of mutually orthogonal parallel lines.The calibration time of this method is about 2S.Compared with the classic tri-orthogonal vanishing point and the calibration toolbox that comes with Matlab,the speed is increased by 231.61%and 387.80%respectively.The 3D reconstruction of the actual calibration object is carried out by this calibration method,and the size error is less than 2.5%.The experimental results show that the underwater calibration method can effectively improve the calibration speed while ensuring the calibration accuracy.It solves the problems of slow underwater calibration speed and harsh calibration restrictions.Based on the research of underwater camera calibration technology based on vanishing point optimization,a set of underwater image calibration system based on vanishing point optimization is designed,and the underwater camera is calibrated by shooting a single image.The specific functions include:(1)Using the method based on the fusion and iterative restoration of ripple characteristics and environmental characteristics to enhance a single underwater image,and output high-quality preprocessed images.(2)Use the fast underwater calibration method based on two sets of mutually orthogonal parallel lines corresponding to the vanishing point optimization to calibrate the underwater camera,and output the internal and external parameters of the camera;(3)Use the calibration error to correct the underwater distortion factor.The work of this paper can provide theoretical and experimental basis for further research on autonomous underwater vehicle target manipulation,marine environment monitoring,underwater cultural relic salvage,underwater augmented reality,marker recognition and other exploration of underwater scenes.