Research On Underwater Vision Measurement Method Based On Deep Learning

With the continuous development of the field of marine resources in my country,the detection of underwater structures,marine organisms,pipelines,etc.puts forward higher requirements for the accuracy of underwater three-dimensional measurement.Sonar has a long detection distance,but its low measurement accuracy cannot meet the measurement requirements.Due to the small attenuation of blue and green lasers,the vision technology using laser scanning has great application prospects underwater.In underwater archaeological work,it can not only display archaeological sites,relics and underwater cultural relics,but also comprehensively archive its spatial information,providing rich data for subsequent further research.This article has carried out research on the three-dimensional measurement method of underwater laser rotation scanning.First,considering the underwater image distortion caused by underwater scattering and refraction,the underwater image is improved by processing the data set through underwater image enhancement and restoration technology.Secondly,considering the interference of the light strip information in the background area when the laser light strip is applied in the underwater environment,the focus is on the light strip detection method based on deep learning,and the convolutional neural network framework based on the dense feature pyramid is built to detect the target light stripe in the image.The ROI area provides more effective data for subsequent calculations.Then the calibration problems of underwater vision measurement system are studied,including axis-eye calibration,underwater camera calibration and underwater light plane calibration.A new axis-eye calibration algorithm is proposed to calculate the conversion relationship between the axis coordinate system and the camera coordinate system.For the process of light refraction in a special underwater environment,an underwater camera imaging model with light refraction is proposed.The internal and external parameters calibration methods of underwater cameras are studied.A high-precision calibration algorithm for underwater light planes is introduced,combined with underwater camera imaging geometric models to complete underwater three-dimensional measurements.Finally,the problem of underwater 3D point cloud registration is studied,and the point cloud ICP registration algorithm is introduced to make the discrete point cloud information integrated,which conforms to human visual perception and highlights the richness of the measured object.Through the underwater three-dimensional measurement of special objects,the experimental results are analyzed from different angles,and the stability of the system and the accuracy of the measurement algorithm proposed in this paper are verified.Through the underwater visual three-dimensional measurement of different objects,the system can meet the basic requirements of measurement.Experimental analysis and data comparison verify the effectiveness of the method.