Enhancing the 3D reconstruction of underwater scenes using sonar and off-the-shelf point cloud denoisers

Side-scan sonar has been extensively used for seabed mapping and SLAM for autonomous underwater vehicles. Its principle consists of scanning the scene with sound waves and analyse the recorded echos to estimate the seabed profile and texture. While the wave propagation, i.e., the data formation process, is generally well understood, the reconstruction of the seabed is still a computationally intensive problem, especially when the area covered is large and/or the scene is observed several times with different viewpoints. The problem becomes even more challenging in the presence of sharp relief changes (e.g., ship wreck, rocks) producing shadowed regions.

In this project, the student will implement a novel, fast algorithm based on existing tools from the computer graphics community to improve and accelerate the seabed reconstruction from sonar data. Traditional image processing tools are usually not well adapted to model non-flat surfaces or can be computationally expensive. Here, the main idea is to couple point cloud denoising tools originally designed to handle efficiently large point clouds (millions of points) with the classical sonar propagation model and produce more accurate 3D maps.

This project is computational and involves modelling aspects as well as implementation of an algorithm using existing toolboxes. The preferred tool used we be Matlab, although it will incorporate C++ subroutines (already implemented).

The interested student is encouraged to contact Dr Altmann prior selecting this project.

Required skills:
-Signal/Image processing
- Experience with Matlab
- Experience C/C++ or python is a plus

Supervisor name: 
Dr. Yoann Altmann
Supervisor and Deputy email addresses: 
Deputy name: 
Prof. Yvan Petillot