Feature-Based Estimation of 3D Clouds with Forward-Looking Imaging Sonar

Bastian Schildknecht tackles the problem of 3D reconstruction of underwater environments by constructing point clouds from sonar images. Having access to accurate maps can often be crucial for many underwater operations. For that reason, a feature-based method for the estimation of landmark positions in 3D coordinates was built. This is done by feature extraction and feature matching on sonar images in order to acquire landmarks. These landmarks then get registered and integrated into the 3D reconstruction by using a novel iterative optimization method for estimating the landmark positions. To verify the performance of the reconstruction method, the self-developed sonar simulator with a synthetic data set is used.