Analysis of Uncertainty in Underwater Multiview Reconstruction

Igor Kozlov
M.S. Thesis Defense

Computer Science

Monday, Aug. 27, 2018, 10:00am
Chase S206

Multiview reconstruction, a method of creating 3D models from multiple images from different views, has been a popular topic of research in the field of computer vision in the last two decades. Increased availability of high-quality cameras led to the development of advanced techniques and algorithms. However, little attention has been paid to multiview reconstruction in underwater conditions. Researchers in a wide variety of fields (e.g., marine biology, archaeology, and geology) could benefit from having 3D models of seafloor and underwater objects. Cameras, designed to operate in air, must be put in protective housings to work with them underwater. This affects the image formation process. The largest source of underwater image distortion results from refraction of light, which occurs when light rays travel through boundaries between media with different refractive indices. This study addresses methods for accounting for light refraction when using a static rig with multiple cameras. We define a set of procedures to achieve optimal underwater reconstruction results, as well as provide analysis of expected quality of the 3D models' measurements.


Igor Kozlov received a bachelor of science in Information Systems and Technologies from Saint Petersburg State Polytechnic University. He is currently pursuing a master’s degree in Computer Science. His research interests include image processing and underwater image reconstruction.