Research Of Underwater Spherical Robot Vision System

In the exploration,mining and protection of underwater resources and energy,or in the fields of underwater archaeological exploration,seabed geomorphology observation,underwater spherical robots have a strong environmental adaptability as a low-cost,high-efficiency machine.It has a wide range of operations,is easy to arm and recycle,and plays a vital role in practical applications.In the military,underwater spherical robots can detect enemy warships at close range and grasp the movements of the enemy forces to achieve an unexpected victory.In terms of civil use,it can replace people's observations of biological observations such as fish in a complex and dangerous environment.This has important theoretical significance and practical application value for China's marine exploration and marine science development.This thesis takes the underwater spherical robot vision system as the research object.And this thesis studies the hardware platform design of the underwater spherical robot vision system,the underwater binocular camera calibration method,the stereo matching method,the underwater image preprocessing method and the underwater target detection segmentation and positioning.The main research work is as follows:Firstly,according to the underwater environment and the characteristics of the platform,the visual hardware system of the underwater spherical robot is selected from the camera to the light source to complete the construction of the visual system hardware platform.Secondly,aiming at the characteristics of the platform of the underwater spherical robot vision system,the specific imaging process of the camera inside the transparent spherical shell is analyzed,and a new underwater multi-media camera model is proposed.The traditional Zhang's calibration method is improved based on the improvement.The nonlinear model calibrates the camera to complete the stereo calibration and correction of the binocular camera,and designs a suitable underwater scene stereo matching algorithm.Finally,we carry out experiments based on the improved calibration method and stereo matching method.Then our experimental result is compared with the traditional Zhang's calibration method in order to prove the effectiveness of the method and the accuracy of the calibration.Then,according to the characteristics of underwater environment imaging,we analysis the principle of land fog imaging.According to the characteristics of underwater light propagation,we make use of the color channel to improve the dark channel prior method and design the underwater image preprocessing method.Finally,we carry out experiments based on this method and analysis the result with other methods to prove this method in the thesis is effective.Finally,we study the target detection segmentation and localization methods of the underwater vision system.In order to segment underwater fish from the background effectively,we design a Mask RCNN network.Using the principle of binocular parallax,the fish in the underwater environment is located by a parallax map.Finally,we carry out experiments and analysis the results which we can prove that our experimental results are consistent with the theory and verify the effectiveness of our method.