| Abstract | Autonomous ships require an awareness system to recognize surrounding obstacles during navigation. In general, the awareness system uses data obtained by sensors such as RADAR (RAdio Detection And Ranging) and AIS (Automatic Identification System). However, autonomous ships operating in the marine environment should recognize small obstacles, including buoys that cannot be detected by RADAR and boats without the AIS system. Just as the tracking accuracy is improved by the fusion of sensors with different characteristics, a camera with a wide-angle of view or a camera capable of night detection can complement each other. In addition, tracking accuracy can be further improved if camera images from multiple viewpoints can be used in the case of detection through multiple ships. Therefore, in this study, camera-based detection and tracking methods were developed to be aware of marine obstacles that could not be recognized by RADAR or AIS. Firstly, the Convolutional Neural Network (CNN) based YOLO v5 model, combined with an attention module, was suggested for obstacle detection. Subsequently, the position and speed of the detected obstacles were tracked using the extended Kalman Filter. Finally, the tracks of the obstacles from various cameras mounted on multiple ships were integrated using the sensor fusion method to increase the accuracy. The field tests were conducted with a USV (Unmanned Surface Vessel) for various ship encountering scenarios to verify the developed methods in the actual marine environment. As a result, it was confirmed that stable and accurate detection and tracking of marine obstacles was performed using multiple cameras. |
|---|---|
| Publication Date | 2022-04-28 |
Jeong-Ho Park, Myung-Il Roh, Hye-Won Lee, Jisang Ha, Yeong-Min Jo, Nam-Sun Son, "Detection and Tracking Methods of Maritime Obstacles Using Multiple Cameras", Proceedings of ICDM(International Conference on Decarbonization and Digitalization in Marine Engineering) 2022, Siheung, Korea, 2022.04.28-29