Contribuições em odometria visual aplicadas em testes não destrutivos

Abstract

Non-destructive testing and evaluation (NDT&E) with ultrasound are essential for evaluating sub-sea components for structural integrity and reliability. Odometry information is an essential feature for NDT&E systems for inspection procedures. Those involve mechanical scanning of the pipeline surface. The position information during the inspection allows for the spatial localization of the transducer and flaws. A simple solution is a mini-wheel quadrature odometer that only measures the testing direction axis. For 2D sweeping, another solution would be to arrange two odometers, one for each axis. Visual odometry techniques were based on cameras operating in different topologies and image processing to overperform the two-dimensional mechanical odometers approach to estimate the position. In underwater environments with depth cameras, the presence of distortions in the 3D visual information due to refraction can significantly impact the position estimation accuracy. Regression or interpolation methods can be applied to model the inverse distortions and are used to correct the distorted points. This study focuses on using odometry and visual odometry techniques for contact and underwater ultrasound NDT, specifically exploring position and direction estimation and correction of underwater distortions by applying corrections models. We study the background theory of applying phase correlation methods for a setup using a monocular camera and inertial unit to estimate the position and attitude of an object. Another subject studied is RGB-D cameras to estimate 3D positions and correct the depth map distortions. The results obtained validate the cameras as a visual odometry tool and shows the possibility of performing a 3D estimation after a corrected depth map.