Reconstrução 3D virtual de componentes reais com dados térmicos para inspeção e manutenção

Abstract

The last decade improvements in the area of Computer Vision, in terms of both hardware and software, promoted and expansion in Virtual Reality and cameras Simultaneous Localization and Mapping (SLAM) in the environment, which brings benefits in several areas, including equipment maintenance. In the Electrical Engineering sector case, many flaws result in the equipment temperature raise, which can be inspected by thermal images, although environment influence and image angle of capture may result in wrong information interpretation. This work proposes a system that aims to achieve these two functions: develop the 3 dimensions real environment reconstruction in the virtual world from captured images, which takes the analysis of the external environment characteristics to the office; and project on these 3D models the temperature obtained by a thermal camera, in order to analyze from many angles the equipment state regarding the presence of any faults. Fo that purpose, a pair of stereo cameras and a thermal camera were aligned and synchronized for image capture during the inspection. Every 3D point cloud obtained is obtained via the applied stereo vision algorithm. For cloud accumulation, or registration, a Visual Odometry technique is applied to gather camera pose from stereo image pair characteristics, followed by filters and overlap between clouds analysis. At the end, for each cloud synchronized with the thermal image, computer vision techniques are used to project the 3D points in the last one, and therefore obtain the temperature data observed. Putting all those pieces together, developed in the ROS framework in C++, a visual and thermal virtual model of the object to be inspected is obtained. The results show the reliability of the applied techniques, aside of models obtained in several scenarios.