基于场景理解的施工临边坠落险兆智能识别方法
Intelligent identification method of construction edge falling near-miss based on scene understanding
韩豫 1李康 2刘泽锋2
作者信息
- 1. 江苏大学 土木工程与力学学院,江苏 镇江 212013;江苏大学 应急管理学院,江苏 镇江 212013
- 2. 江苏大学 土木工程与力学学院,江苏 镇江 212013
- 折叠
摘要
为更及时、更有效地预防施工临边坠落事故的发生,并弥补现有智能预警方法在场景理解方面的不足,融合深度学习与语义推理,提出1 种险兆识别方法.该方法通过neo4j构建险兆知识图谱,将引入轻量级视觉Transformer的YOLOx模型识别工人的险兆行为,设计描述空间关系的IoU计算方法并使用Cypher推理语言进行险兆推理.研究结果表明:施工临边坠落各要素识别的平均精度达91%以上,且IoU计算与险兆推理准确率均为100%,模型识别效果与险兆推理效果较好,该方法总体满足精度和速度的识别要求.研究结果可为实现施工临边坠落险兆行为的精准识别和预警提供参考.
Abstract
In order to prevent the construction edge fall accidents more timely and effectively,and make up for the shortcom-ings of existing intelligent early-warning methods in scene understanding,a near-miss identification method was proposed by integrating deep learning and semantic inference.This method first constructed a near-miss knowledge graph through neo4j,and then introduced the YOLOx model with a lightweight visual Transformer to identify the near-miss behavior of workers.An IoU calculation method for describing the spatial relationship was designed,and the Cypher inference language was used to carry out the near-miss inference.The results show that the average accuracy of identifying various elements of construction edge falling is over 91%,and both the accuracy of IoU calculation and near-miss inference are 100%.The model identifica-tion effect and near-miss inference effect are good.The method generally meets the identification requirements of accuracy and speed.The research results can provide reference for achieving the accurate identification and early-warning of construction edge falling near-miss behavior.
关键词
临边坠落/场景理解/深度学习/知识图谱/险兆推理Key words
edge falling/scene understanding/deep learning/knowledge graph/near-miss inference引用本文复制引用
基金项目
国家自然科学基金(72071097)
教育部人文社会科学研究规划基金(20YJAZH034)
江苏大学应急管理学院专项科研项目(KY-B-10)
2023年江苏省研究生科研创新计划项目(KYCX23_3674)
出版年
2024
NETL
NSTL
万方数据