Underwater 3D Reconstruction Of Fusing Photometric Stereo And Structured Light

The existing underwater 3D reconstruction methods mostly are the methods in air,which obtain low accuracy when applied to underwater directly,because the absorption and scattering of light in water is far greater than that in air.In this thesis,we improve the structured-light reconstruction method in air to improve its reconstruction accuracy in water.Then,a height correction algorithm is proposed for integral surface in photometric stereo to reduce deformation of integral surface.Finally,a method that fuses photometric stereo and structured light is proposed.The method makes up for the advantages and disadvantages of the two reconstruction results,which makes the final reconstruction surface has both high-precision height information and gradient information.It mainly includes the following contents:(1)For structured light reconstruction method in air,we proposed an improved method to improve the reconstruction accuracy in water.In the pre-processing stage of structured light image,we added backscatter removed process.Based on the underwater imaging model,the structured light image is removed backscatter from the perspective of image restoration.In the extraction stage of feature points in structured light strip image,we combinate maximum value method and image erosion,which improves robustness of the method in turbid water.In height reconstruction stage of feature points,for the results of structured light reconstruction,we use gradient constraints to distinguish normal points and abnormal points,and use GRNN to regenerate the height value of the abnormal point.Finally,the reconstruction accuracy and the robustness to different water quality are improved greatly.(2)For the problem that the integral surface in photometric stereo has large height error,we propose a height correction algorithm to reduce deformation of integral surface.Frstly,four integral methods(Frankot-Chellappa method,Poisson method,Diffusion method and M-estimators method)were evaluated,and the integral surface obtained by the integral method with better evaluation results was corrected to improve the gradient integral effect.(3)Based on results of structured light reconstruction and photometric reconstruction,the method fuses structured light and photometric stereo is proposed.We use BP neural network to establish height mapping of photometric stereo to structured light,that is,height of integral surface in photometric stereo is corrected by height of structured light.Our method improves the height accuracy of integral surface while maintaining its high accuracy of gradient.The experiment results show that the underwater 3D reconstruction method that fuses photometric stereo and structured light can make the final reconstructed surface have high-precision gradient information and height information.What's more,the fusion method is superior to structured light in height information,and is superior to photometric stereo in gradient information.The fusion method has applicability to different target objects and is robust to water quality,which can be applied to the fields of mariculture industry,seabed topography,underwater archaeology and seabed resource development.