Peiru Shi , Miao Gao , Shuai Chen , Qianfeng Jing , Yu Xia , Yu Han , Anmin Zhang
{"title":"基于船舶姿态估计和COLREGs量化的大规模智能避碰决策","authors":"Peiru Shi , Miao Gao , Shuai Chen , Qianfeng Jing , Yu Xia , Yu Han , Anmin Zhang","doi":"10.1016/j.oceaneng.2025.121679","DOIUrl":null,"url":null,"abstract":"<div><div>Collision avoidance decision-making is crucial for the autonomous navigation of Maritime Autonomous Surface Ships (MASS). Visual pose estimation and navigation light signals provide essential information for evaluating encounter situations under the “in sight” condition defined by COLREGs. This study proposes an intelligent collision avoidance method based on ship pose estimation (SPE) and COLREGs quantification. First, the annotation specifications for the SPE dataset are formulated, and the dataset is constructed accordingly. Then, by optimizing the network structure and loss function of YOLO11m-pose, we develop the GEW-YOLO architecture and train it on the SPE dataset. Subsequently, pose estimation results are used to classify encounter situations and guide decision-making based on encounter characteristics and historical data. In addition, ablation experiments were conducted to validate the proposed GEW-YOLO model, showing a 20.2 % reduction in parameters, a 13.2 % decrease in GFLOPs, and improvements of 5.2 % and 4.7 % in mAP50-95 for detection and pose estimation compared to the baseline. Finally, collision avoidance experiments conducted on self-developed MASS confirm that the proposed method can make accurate collision avoidance decisions across various encounter situations.</div></div>","PeriodicalId":19403,"journal":{"name":"Ocean Engineering","volume":"335 ","pages":"Article 121679"},"PeriodicalIF":5.5000,"publicationDate":"2025-06-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"MASS intelligent collision avoidance decision-making based on ship pose estimation and COLREGs quantification\",\"authors\":\"Peiru Shi , Miao Gao , Shuai Chen , Qianfeng Jing , Yu Xia , Yu Han , Anmin Zhang\",\"doi\":\"10.1016/j.oceaneng.2025.121679\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<div><div>Collision avoidance decision-making is crucial for the autonomous navigation of Maritime Autonomous Surface Ships (MASS). Visual pose estimation and navigation light signals provide essential information for evaluating encounter situations under the “in sight” condition defined by COLREGs. This study proposes an intelligent collision avoidance method based on ship pose estimation (SPE) and COLREGs quantification. First, the annotation specifications for the SPE dataset are formulated, and the dataset is constructed accordingly. Then, by optimizing the network structure and loss function of YOLO11m-pose, we develop the GEW-YOLO architecture and train it on the SPE dataset. Subsequently, pose estimation results are used to classify encounter situations and guide decision-making based on encounter characteristics and historical data. In addition, ablation experiments were conducted to validate the proposed GEW-YOLO model, showing a 20.2 % reduction in parameters, a 13.2 % decrease in GFLOPs, and improvements of 5.2 % and 4.7 % in mAP50-95 for detection and pose estimation compared to the baseline. Finally, collision avoidance experiments conducted on self-developed MASS confirm that the proposed method can make accurate collision avoidance decisions across various encounter situations.</div></div>\",\"PeriodicalId\":19403,\"journal\":{\"name\":\"Ocean Engineering\",\"volume\":\"335 \",\"pages\":\"Article 121679\"},\"PeriodicalIF\":5.5000,\"publicationDate\":\"2025-06-07\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Ocean Engineering\",\"FirstCategoryId\":\"5\",\"ListUrlMain\":\"https://www.sciencedirect.com/science/article/pii/S002980182501385X\",\"RegionNum\":2,\"RegionCategory\":\"工程技术\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q1\",\"JCRName\":\"ENGINEERING, CIVIL\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Ocean Engineering","FirstCategoryId":"5","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S002980182501385X","RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"ENGINEERING, CIVIL","Score":null,"Total":0}
MASS intelligent collision avoidance decision-making based on ship pose estimation and COLREGs quantification
Collision avoidance decision-making is crucial for the autonomous navigation of Maritime Autonomous Surface Ships (MASS). Visual pose estimation and navigation light signals provide essential information for evaluating encounter situations under the “in sight” condition defined by COLREGs. This study proposes an intelligent collision avoidance method based on ship pose estimation (SPE) and COLREGs quantification. First, the annotation specifications for the SPE dataset are formulated, and the dataset is constructed accordingly. Then, by optimizing the network structure and loss function of YOLO11m-pose, we develop the GEW-YOLO architecture and train it on the SPE dataset. Subsequently, pose estimation results are used to classify encounter situations and guide decision-making based on encounter characteristics and historical data. In addition, ablation experiments were conducted to validate the proposed GEW-YOLO model, showing a 20.2 % reduction in parameters, a 13.2 % decrease in GFLOPs, and improvements of 5.2 % and 4.7 % in mAP50-95 for detection and pose estimation compared to the baseline. Finally, collision avoidance experiments conducted on self-developed MASS confirm that the proposed method can make accurate collision avoidance decisions across various encounter situations.
期刊介绍:
Ocean Engineering provides a medium for the publication of original research and development work in the field of ocean engineering. Ocean Engineering seeks papers in the following topics.