| The demand for accurate 3D reconstruction technology at absolute scale under deep ocean is continuously expanding,as ocean observation technology keeps exploring the boundaries.Meanwhile,certain progress on computer visualization field allows the binocular 3D reconstruction and measurement theory,previously proven in atmosphere only,gradually applicable in underwater condition now.However,the current underwater 3D reconstruction and measurement technology had few considerations on underwater applications and effects caused by refraction,or sometimes simply compensated by distortion.This thesis obtains more accurate reconstruction results by conducting a specific study on the light refraction path underwater.This thesis mainly focused on the non-single-view camera model based on the coplanar constraint of light,which simulates the process of underwater imaging and realizes the 3D reconstruction under absolute scale of the underwater target.The study experienced through the restoration for underwater image,image feature extraction and matching,image stereo matching,target head-and-tail point selection based on convolutional neural network,etc.,thus completed the technical framework and implementation procedure for 3D reconstruction considering underwater refraction and measurement under absolute scale in the final stage.The thesis described a design and progress of an experiment for the camera with waterproof cover,shooting the underwater objects in the water tank by using the binocular camera outside the glass water tank to simulate the process of shooting the underwater scene.Both single-view and non-single-view camera model were applied to shoot the fish model underwater in the 3D reconstruction experiment.The experiment used the coplanar constraint of the two refracted rays to calculate the distance from the optical center to the glass and the thickness of the glass to directly connect the underwater 3D point cloud with the 2D image and a new virtual optical center,and project the original pixel image into the three-dimensional space with the new virtual optical center to obtain the dense point cloud of the object,thus concluded the three-dimensional reconstruction result combined with the camera pose calculation.In the object length measurement experiment,the target object area was binarized to find the minimum circumscribed moment of the target object,and to calculate the intersection of the target object and the two short sides of the minimum circumscribed rectangle.After that,according to the camera projection model,the intersection point is projected to the world coordinate system,thus the length of the object could be calculated based on Euclid geometry.The experimental results proved that the method model described in this thesis is effective and can meet the requirements for better measurement accuracy.Besides,this thesis conducted a 3D reconstruction experiment on the seabed data captured by the underwater binocular camera based on the built 3D reconstruction framework.A restoration was performed with underwater image imaging model on the dataset.Also,the SIFT operator was used to extract and match feature points on the sequence images,and these feature points are projected into the three-dimensional space.After the eight-point method and re-projection error optimization applied to obtain the pose of the camera,the depth map of the image was calculated using the stereo matching method,and the sequence images were projected into the three-dimensional space according to the depth map to obtain a dense three-dimensional point cloud.This thesis also conducts 3D reconstruction and measurement experiments on the wave scene in the offshore area.Firstly,the binocular camera is calibrated,then the stereo matching method was used to calculate the depth map and the image was projected into the three-dimensional space according to the depth map to form a dense three-dimensional point cloud.Finally,the point cloud was converted into an elevation map measured by height in metric units. |
PDF Full Text Request