Chapter Four - Autonomous environment and target perception of underwater offshore vehicles

Abstract

An autonomous underwater vehicle (AUV) relies on an underwater camera and sonar for perception the surrounding environment including underwater image processing, object detecting and tracking. For underwater optical environment perception, the limited visual range of the camera severely limits the acceptance of detection information; in addition, underwater light is assimilated and scattered, which seriously deteriorate the underwater imaging, especially decreases the image contrast. To address these issues, a joint framework based on CNN is proposed to improve underwater image quality and extract more matching feature points. For accurate registration between images, an underwater mosaicking technology is also involved, and a fusion algorithm is implemented to mitigate artificial mosaicking traces. Experimental results show that our presented framework can not only keep underwater image detail information, but also exact more matching feature points for registration and mosaicking. For underwater acoustic image environment perception, mechanically scanned imaging sonars (MSIS) are usually equipped on AUVs for avoiding obstacles and multiple-targets tracking. In this chapter, two multiple underwater object tracking methods are presented. The proposed methods are using the cloud-like model data association. Some sea trials are implemented to validate the effectiveness of the presented algorithms. Experiment results demonstrate that the presented cloud-like model data association method has the characteristics of more accurate clustering. Multiple targets were finally clustered for AUV stable tracking.