RIASSUNTO
Divers have a variety of tasks typically with high complexity and high risk. Autonomous underwater vehicles (AUVs) can improve the efficiency and safety of divers by assisting divers in performing underwater operations. A dive-buddy AUV should have the ability to follow a diver and interact with a diver. Dive-buddy AUVs can be developed based on machine vision technology or acoustics technology, leading to different advantages and limitations regarding the diver-following and human-machine interaction abilities of the AUVs. Compared with sonar devices, the main limitation of optical cameras is their short underwater visibility range. However, optical cameras have the advantages of high resolution, high frame rate, low cost, and high application popularity. The IUT AUV, developed by the Institute of Undersea Technology at Nation Sun Yat-sen University, is the testbed AUV of this research. This research aimed to make the IUT AUV become a diver-following AUV. This research applied Tiny-YOLOv3 convolutional neural network (CNN) to image detection of divers and developed a diver detection module as a payload sensor for the IUT AUV. This research developed a single-diver following control algorithm and evaluated the single-diver following performance under different scenarios through hardware-in-the-loop simulation (HIL) system that conforms to the communication format and power architecture of the IUT AUV. This research verified the single-diver following capability of the IUT AUV through closed water experiments conducted in a towing tank.