Underwater Stereo Processing

A project of the Robotics 2019 class of the School of Information Science and Technology (SIST) of ShanghaiTech University. Course Instructor: Prof. Sören Schwertfeger.

Conghui Zhang, Deming Wu, Tianxing Niu

Omni-directional cameras provide a comprehensive view of the scene in the computer vision and SLAM tasks, which make it receiving more and more attention. However, the use of these cameras underwater is still at a very early technological stage. This project manages to apply SGBM algorithm to complete the underwater omni-directional mapping by stereo camera video. The frames are unwrapped into images with the spherical model, and preprocessing is necessary to obtain better results. The rotation matrix from calibration can work out the position difference between the two cameras, which geometic calculation method also leads to the same, and the difference is used to align two images. The new pair is inputed into SGBM to get desparity maps. Finally, desparity maps of two same-time frames in different occasions are acquired in our paper, where obvious boundaries are easily distinguished.

Left and right frame from the underwater omni stereo camera

Video of one input camera: