Thesis - Jeong-Ho Park, "A Method for Detection and Tracking of Maritime Obstacles Based on Multi-Video", M.Sc. Thesis, Seoul National University, 2023.02.24

Extra Form
Abstract	Among the causes of marine accidents, human error accounts for a relatively high rate, and accordingly, the need for an autonomous recognition technology for recognizing the surroundings is emerging. Research on autonomous recognition technology using traditional recognition sensors such as Automatic Identification Systems (AIS) and Radio Detection and Ranging (RADAR) is being actively conducted, but there are clear limits. Therefore, we tried to develop a new cognitive technology to replace them. In this paper, we proposed an autonomous recognition technology using a camera to supplement the limitations of traditional cognitive sensors and replace human vision. First, the YOLOv5 algorithm, a real-time object detection algorithm based on camera images, was improved to increase obstacle detection accuracy. Then, a position transformation algorithm estimated the relative position of the detected obstacle. Based on the relative position of the obstacle, we proposed an adaptive extended Kalman filter to estimate the motion of the obstacle, such as trajectory, Course Over Ground (COG), and Speed Over Ground (SOG). In addition, assuming that USV is operated for strategic purposes and mutual communication and cooperation are possible, sensor fusion between data tracked by different cameras was performed to increase the accuracy of tracking data. It was confirmed that more accurate tracking data could be obtained by fusing the tracked data from several cameras to improve tracking accuracy or compensate for the disadvantages that occur in the tracking process of individual cameras.
Publication Date	2023-02-24

Abstract

Among the causes of marine accidents, human error accounts for a relatively high rate, and accordingly, the need for an autonomous recognition technology for recognizing the surroundings is emerging. Research on autonomous recognition technology using traditional recognition sensors such as Automatic Identification Systems (AIS) and Radio Detection and Ranging (RADAR) is being actively conducted, but there are clear limits. Therefore, we tried to develop a new cognitive technology to replace them.
In this paper, we proposed an autonomous recognition technology using a camera to supplement the limitations of traditional cognitive sensors and replace human vision. First, the YOLOv5 algorithm, a real-time object detection algorithm based on camera images, was improved to increase obstacle detection accuracy. Then, a position transformation algorithm estimated the relative position of the detected obstacle. Based on the relative position of the obstacle, we proposed an adaptive extended Kalman filter to estimate the motion of the obstacle, such as trajectory, Course Over Ground (COG), and Speed Over Ground (SOG). In addition, assuming that USV is operated for strategic purposes and mutual communication and cooperation are possible, sensor fusion between data tracked by different cameras was performed to increase the accuracy of tracking data. It was confirmed that more accurate tracking data could be obtained by fusing the tracked data from several cameras to improve tracking accuracy or compensate for the disadvantages that occur in the tracking process of individual cameras.

Publication Date

2023-02-24

Jeong-Ho Park, "A Method for Detection and Tracking of Maritime Obstacles Based on Multi-Video", M.Sc. Thesis, Seoul National University, 2023.02.24

박정호, "다중 영상 기반 해상 장애물 탐지 및 추적 방법", 석사학위논문, 서울대학교, 2023.02.24

Jeong-Ho Park, "A Method for Detection and Tracking of Maritime Obstacles Based on Multi-Video", M.Sc. Thesis, Seoul National University, 2023.02.24

단축키

Articles