| Abstract | Situational awareness is one of the most essential abilities of unmanned vessels. Even in the case of manned vessels, situational awareness can contribute to safe navigation by detecting and alerting potential collisions. Although radar and AIS(Automatic Identification System) are widely used for detection, it is necessary to use vision cameras that can take place of human eyes to detect near objects and identify object types. In this study, we performed machine vision based object detection and tracking for the situational awareness in maritime environment. For object detection, the state-of-the-art detection algorithms and their various backbone CNN(Convolutional Neural Network) models were applied; a two-stage detection model derived from Faster R-CNN and a single-stage detection model based on YOLO were implemented and tested in this study. The performance in mAP(mean average precision) score of each detection model was evaluated and compared. For object tracking, we surveyed not only conventional correlation filtering algorithms but also deep learning algorithms using LSTM(Long Short-Term Memory) network models. All the trainable detection and tracking models were trained by maritime domain image dataset. Performance of each model was estimated under maritime visionary environment. |
|---|---|
| Publication Date | 2019-09-24 |
Sung-Jun Lee, Myung-Il Roh, Min-Jae Oh, Youngsoo Seok, Won-Jae Lee, June-Beom Lee, Hyun Soo Kim, "Image-based Object Detection and Tracking Method for Ship Navigation," Proceedings of ICCAS(International Conference on Computer Applications in Shipbuilding) 2019, Rotterdam, Netherlands, pp. 89-92, 2019.09.24-26