Reconstruction of 3D Points From Uncalibrated Underwater Video

3D Reconstruction;Computer Vision;Autocalibration;System Design Engineering

Cavan, Neil

University of Waterloo

关键词: 3D Reconstruction;    Computer Vision;    Autocalibration;    System Design Engineering;   

Others  :  https://uwspace.uwaterloo.ca/bitstream/10012/6242/1/Cavan_Neil.pdf

瑞士|英语

来源: UWSPACE Waterloo Institutional Repository

【 摘 要 】

This thesis presents a 3D reconstruction software pipeline that is capable of generatingpoint cloud data from uncalibrated underwater video. This research project was undertakenas a partnership with 2G Robotics, and the pipeline described in this thesis will becomethe 3D reconstruction engine for a software product that can generate photo-realistic 3Dmodels from underwater video. The pipeline proceeds in three stages: video tracking,projective reconstruction, and autocalibration.Video tracking serves two functions: tracking recognizable feature points, as well as selecting well-spacedkeyframes with a wide enough baseline to be used in the reconstruction. Video tracking is accomplishedusing Lucas-Kanade optical flow as implemented in the OpenCV toolkit. This simple andwidely used method is well-suited to underwater video, which is taken by carefully pilotedand slow-moving underwater vehicles.Projective reconstruction is the process of simultaneously calculating the motion of thecameras and the 3D location of observed points in the scene. This is accomplished usinga geometric three-view technique. Results are presentedshowing that the projective reconstruction algorithm detailed here compares favourably tostate-of-the-art methods.Autocalibration is the process of transforming a projective reconstruction, which is notsuitable for visualization or measurement, into a metric space where it can be used. Thisis the most challenging part of the 3D reconstruction pipeline, and this thesis presents anovel autocalibration algorithm. Results are shown for two existing cost function-basedmethods in the literature which failed when applied to underwater video, as well as theproposed hybrid method. The hybrid method combines the best parts of its two parentmethods, and produces good results on underwater video.Final results are shown for the 3D reconstruction pipeline operating on short under-water video sequences to produce visually accurate 3D point clouds of the scene, suitablefor photorealistic rendering. Although further work remains to extend and improve thepipeline for operation on longer sequences, this thesis presents a proof-of-concept methodfor 3D reconstruction from uncalibrated underwater video.