Sea Surface Object Detection

• Main Authors: Jun-Hao Huang, 黃俊豪
• Other Authors: 丁肇隆
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/6e9cg4

碩士 === 國立臺灣大學 === 工程科學及海洋工程學研究所 === 106 === With the rapid development of deep learning, autopilot is one of the most popular research projects in recent years. Whether autopilot car or drone have a very good research result and the actual product in the real world has a great success. It seems that autopilot is a future trend and we supposed that this technical can do a well job in autopilot assistant system. Modern ships are often equipped with various equipment to assist navigation, for example, radar, sonar, etc. In addition to navigating, the most important system of the ship is to avoid collision at sea. A real-time detected system can avoid ship in danger. However, if the size of the object is not enough to be detected by the radar, human still rely on their vision to detect ships or obstacles on sea surface. Therefore, we believe that computer vision can become a reliable assistant system for navigation. In recent years, Convolutional Neural Network (CNN) has achieved great success in computer vision, image recognition, and other applications. CNN provides powerful computing ability and its deep architecture also provides reliable test results. In this research, we proposed a sea surface object detection system using Fully Convolutional Network (FCN) and imported stereo vision to estimate the distance and azimuth of the object to achieving an assistant system for navigation.