이수민, 노명일, 하지상, 이원재, “해상 이미지로부터 선박의 위치를 추정하는 방법”, 한국CDE학회 논문집, Vol. 25, No. 3, pp. 320-328, 2020.09.01
조회 수 217
|Recently, research on smart ships is actively underway. One of the key technologies for smart ships is ANS (Autonomous Navigation System). To ensure the reliability of such ANS, it is important to estimate the locations of other ships. Although the method of utilizing the AIS (Automatic Identification Information) or RADAR (RAdio Detection And Ranging) is very efficient, it is difficult to guarantee the reliability of the ANS depending on only the equipment since small boats often operate without them. This problem can be solved by estimating the locations of other ships from ocean images using cameras on the own ship. To estimate the locations from the images, the horizon must be detected. The accuracy of the existing method for detecting the horizon is very low when the weather is cloudy, or there are obstacles on the sea. In this study, 300 image data sets are generated using the Singapore Maritime Dataset to train a deep learning model called the Fully Convolutional Network (FCN) for detecting the horizon. Using the trained model, the horizons of the actually captured images are detected. Then, the locations of other ships are calculated using the detected horizon, and the results are compared with the AIS data. The results show that the proposed method can detect the horizon and can be used to estimate the locations of other ships from ocean images.