Fusing Depth And Multispectral Photometric Stereo For 3D Reconstruction And Its Applications In Underwater Environment

Reconstruction of the 3D geometry of objects from their 2D images is the basic problem in computer vision, which can be applied in the fields such as industrial measurement, human-computer interaction, medicine, archaeology, computer animation, et al.Current theories and methods in 3D reconstruction were developed under the assumption of the air environment. However, the underwater environment is much more complicated. The absorption and scattering effect caused by the suspended particles attenuate the image signal, where the reconstruction algorithms failed under the water.This thesis uses depth sensor to obtain the objects’ depth data, then uses multispectral photometric stereo to calculate the objects’ surface normal. Those two part are fused for 3D reconstruction. At the same time, this thesis simplify the under water image formation model, to make it apply to our method.This thesis implements a multispectral photometric stereo 3D reconstruction technology based on depth sensor. The object was illuminated with three light sources of the color:red, green, and blue, simultaneously. Assume that an object’s surface reflection complies with lambert model, partitioning the surface area based on color, then the color images are used to compute the normal vector of the object surface.For the different effects of absorption and scattering on different spectrum light, this paper improve the underwater imaging model, and introducing the method explained in the previous paragraph to the underwater environment. Experiments on different turbidity of water environment show that the method of this thesis can reconstruct under water object in a more simple operation, cheap equipment and low amount of calculation.