Preprocessing and Future Extraction for Visual Underwater Odometry

Abstract

Distortion of underwater images can impair the accuracy and robustness of visual odometry algorithms, leading, in particular, to a decrease in the number of detected conjugate key points on pairs of images. In this regard, image preprocessing, the selection of key points and the odometry algorithms themselves are important questions in solving underwater robotics problems. The paper considers on real data the influence of various algorithms for preprocessing and constructing key points on the properties of video odometry algorithms in conditions of underwater and uncontrolled camera movement. It is shown that the number of key points determined on the sequence of images, and, consequently, the accuracy of obtaining the trajectory of the camera movement can depend radically on both preprocessing and the detector used.