船员不安全行为特征数据集

船员不安全行为特征数据库管理软件在研发过程中，针对船员未穿戴安全帽和救生衣、船员疲劳、驾驶员离岗、船员抽烟和打电话以及违规操作缆绳等不安全行为，建立了船员不安全行为特征数据集，其中，安全帽、救生衣数据集由9547张图片组成，疲劳检测数据集由46511张图片组成，人员跌倒数据集由4666张图片组成，人员离岗数据集由3219张图片组成，吸烟、打电话数据集由10674张图片组成，缆绳违规操作数据集由5300张图片组成。针对所建立的数据集，设计了三种基于计算机视觉的不安全行为与状态辨识算法，分别为船上多尺度目标检测模型、船员疲劳监测模型以及缆绳违规操作识别模型。通过在建立的船员不安全行为特征数据集上的验证，6类不安全行为的识别准确度均能达到94%以上，且在实船场景下具有较好的泛化性，并通过了第三方测试。

During the development of the database management software for crew unsafe behavior characteristics, a crew unsafe behavior characteristic dataset was established targeting various unsafe behaviors including crew members not wearing safety helmets and life jackets, crew fatigue, driver's absence from post, crew smoking and phone calling, and illegal cable operation. Specifically, the safety helmet and life jacket dataset comprises 9547 images, the fatigue detection dataset consists of 46511 images, the person fall dataset has 4666 images, the person absence from post dataset contains 3219 images, the smoking and phone calling dataset is made up of 10674 images, and the illegal cable operation dataset includes 5300 images. Based on the established dataset, three computer vision-based algorithms for unsafe behavior and state recognition were designed, namely the on-board multi-scale object detection model, crew fatigue monitoring model, and illegal cable operation recognition model. Validation experiments conducted on the established crew unsafe behavior characteristic dataset demonstrate that the recognition accuracy of all 6 types of unsafe behaviors exceeds 94%, and the models exhibit good generalization performance in real ship scenarios, having passed third-party testing.

搜集汇总

背景与挑战

背景概述

该数据集由大连海事大学创建，专注于船员不安全行为的计算机视觉识别，包含安全帽和救生衣穿戴、疲劳、跌倒、离岗、吸烟打电话及缆绳违规六类行为的图片数据，总计超过8万张图片。数据集支持开发了三种辨识算法，验证识别准确度达94%以上，并具备实船场景泛化能力，源自国家重点研发计划项目'在航船舶安全风险辨识与防控平台'，用于提升海上运输安全风险管理。