3D Image from Sensor Fusion

Critical missions performed by the US Navy require accurate and efficient classification of objects in underwater optical video. This task is performed by a trained operator; however, the optical data is generally of poor quality (low contrast and/or low illumination), and when performed during missions the real-time video stream cannot be paused or rewound. These challenges can cause high-confidence object classification to take a considerably longer time than is desirable. On this effort, Toyon Research Corporation proposes to develop efficient algorithms implemented in a software application tool that will perform accurate 3D reconstruction of objects in underwater video. By leveraging and optimizing novel in situ sensor calibration algorithms, incremental sparse bundle adjustment, and efficient 3D reconstruction algorithms, the developed software will form accurate 3D models, display the models to the operator, and update the models as optical data from varying perspectives is collected. Furthermore, the algorithms will fuse the optical data with existing sonar data for increased accuracy and efficiency.