Calibration Of Underwater Dome Port Camera Systems And The Application To Deep Sea Gas Flow Quantification

With the rapid development of photogrammetric computer vision technology and a growing interest in the deep sea environment by ocean science,high-quality underwater camera systems have been extensively used as visual sensors for the surveying and exploration of the huge uncharted terrain on Earth-deep sea.In order to protect the cameras from water and extremely high pressure,the cameras are concealed in a pressureresistant housing,observing the world through a glass window-a flat port or a dome port.The light rays change their direction at the glass interface when entering the camera sensor since the media is different(water-glass-air).This introduces the geometric distortion on the images and leads to a systematic error when the standard on-land computer vision algorithms are directly applied to the underwater scenarios.It is necessary to discover the special camera model when a perspective camera sitting in a spherical glass port,and also investigate its corresponding calibration approaches for many machine vision applications.In recent years,as the leakage of greenhouse gases from the seafloor has exacerbated global warming,quantitatively measuring the amount of gas emitted from the seafloor has become an important topic in marine research.Since the gases,mostly carbon dioxide and methane,released from the seafloor into the water column as rising bubbles,can be photographed by high-speed cameras into video sequences.The image processing techniques can be employed to extract the bubble information from images and measure the amount of gas released.Currently,the stereo vision gas release monitoring algorithms pose many issues that are not well handled,such as camera calibration,bubble matching across stereo image pairs,temporal bubble tracking,etc.Therefore,it is important to investigate those issues to improve the accuracy and robustness when measuring the gas flow.This thesis mainly focuses on two aspects:(1)First,the spherical refraction model and the camera geometry when the perspective camera center is deviated from the glass port center is studied,then the refraction center,the refraction center co-linear constraint are proposed.After that,a 8-point algorithm is proposed to estimate the refraction center.Finally,two approaches are proposed to adjust and calibrate the dome port camera system.(2)Focusing on the key issues in stereo vision ocean gas release monitoring algorithm,the stereo vision system for the gas measurement device is adjusted and calibrated,then a temporal median filter technique is employed to remove the background in the bubble images,then the epipolar geometry constraint is used as matching reference to find bubble correspondences in the image pairs.Once the bubble correspondences are determined,the ellipsoid triangulation technique is used to calculate volume of each individual bubble.Finally,the Kalman framework is used to model the motion of the bubble and to track the bubbles overtime.The synthetic experiments and real-world experiments are conducted to evaluate the correctness and effectiveness of the approaches proposed in this thesis.