Yeong-Min Jo, Myung-Il Roh, Hye-Won Lee, Jisang Ha, Do-Hyun Chun, Min-Chul Kong, "An Improved Method for the Sensor Fusion for Autonomous Ships", Proceedings of 10th PAAMES and AMEC 2023, Kyoto, Japan, 2023.10.18-20
International Conference
2023.10.23 10:46
Yeong-min Jo, Myung-Il Roh, Hye-Won Lee, Jisang Ha, Do-Hyun Chun, Min-chul Kong, "An Improved Method for the Sensor Fusion for Autonomous Ships", Proceedings of 10th PAAMES and AMEC 2023, Kyoto, Japan, 2023.10.18-20
조회 수 132
| 첨부 '1' |
|---|
| Abstract | To avoid obstacles in an autonomous ship, it is necessary to track the exact location of each obstacle. For this purpose, sensors such as AIS (Automatic Identification System), RADAR (RAdio Detection And Ranging), and cameras are used for detecting other obstacles. Since each sensor has different characteristics, such as error distribution, detection period, and blind spot, it is necessary to use each sensor’s data simultaneously to supplement each other to detect obstacles accurately. This is where sensor fusion methods play a critical role. The sensor fusion method estimates the exact position, speed, and direction of the obstacles based on the state information of the obstacles tracked through different sensor data. In the previous study, we applied one of the sensor fusion methods, which fused the sensor data with the weights calculated from the error distribution. However, in the method, some of the sensor characteristics were difficult to be reflected on tracking results and did not affect the weights of the sensor fusion method. To address this issue, we implemented various methods for sensor fusion and compared their performance. To evaluate the effectiveness of the sensor fusion methods, a ship navigation situation was implemented in a virtual environment similar to reality. We implemented a multiple object tracking situation in the virtual environment to see if we could simultaneously track and fuse different obstacles well. Three types of sensors were used in the environment: AIS, RADAR, and camera. Tracking and fusion were performed by generating sensor data virtually. Each fusion method was applied based on a track-level fusion structure, which fuses tracking results for each sensor data. The effectiveness of each fusion method was confirmed by comparing the fusion result with the actual position of the objects to be detected. It was confirmed that most of the methods were improved compared to the fusion results in the previous study. |
|---|---|
| Publication Date | 2023-10-20 |
