基于大场景视频监控的坝面施工机械潜在碰撞风险预警方法
An early warning method for potential collision risk of dam surface construction machinery based on large scene video surveillance
曾拓程 1王佳俊 1钟登华 2张雨诺 1康栋3
作者信息
- 1. 天津大学水利工程智能建设与运维全国重点实验室,天津 300350
- 2. 天津大学水利工程智能建设与运维全国重点实验室,天津 300350;中国农业大学水利与土木工程学院,北京 100083
- 3. 雅砻江流域水电开发有限公司,四川成都 610051
- 折叠
摘要
基于大场景视频监控实现坝面施工机械潜在碰撞风险预警对保证大坝施工安全具有重要意义.然而,目前坝面施工机械潜在碰撞风险检测主要依赖人工经验判断,易出现漏判和误判等问题.因此,本研究提出一种基于大场景视频监控的坝面施工机械潜在碰撞风险预警方法.首先,基于Trajectron++轨迹预测算法,通过迁移学习,实现对坝面大场景视频监控中数量众多、类型多样的施工机械在未来一段时间内的轨迹预测.其次,提出将行驶接近时间和机械最大拥挤度作为坝面施工机械潜在碰撞风险的量化指标,并基于模糊规则,建立不同行驶速度条件下两个指标与潜在碰撞风险分级预警的模糊隶属度函数.最后,采用证据理论对两个指标的预警结果进行融合,计算最终的预警等级.以两河口大坝施工现场的大场景视频监控为例进行实验验证,结果表明坝面施工机械未来6 s的轨迹预测平均位移误差和最终位移误差分别为1.17和2.36 m,且基于模糊-证据融合的施工机械潜在碰撞风险分级预警结果可为坝面施工安全提供自动化、智能化分析方法.
Abstract
Based on large-scene video monitoring,it is important to realize the potential collision risk warning of dam construction machinery to ensure the safety of dam surface construction.However,at present,it mainly re-lies on manual experience to judge the potential collision risk of dam surface construction machinery,which is prone to problems such as omission and misjudgment.Therefore,this paper proposes an early warning method for potential collision risk of dam surface construction machinery based on large-scene video surveillance.First,based on Trajectron++algorithm,the trajectory prediction of multi-type construction machinery with large quanti-ty in the future period of time in the large-scene video surveillance of the dam surface is realized through transfer learning;then,the drive proximity time and maximum crowdedness of machinery are proposed as the quantitative indexes of the potential collision risk of the construction machinery on the dam surface,and the fuzzy membership function of potential collision risk graded warning under different driving speeds of construction machinery is estab-lished based on the fuzzy rules;finally,the warning results of the two indicators are fused with evidence fusion theory to obtain the final warning level.The proposed method is tested on the large-scene video surveillance of the Lianghekou dam construction site.The average displacement error and the final displacement error of the trajectory prediction of the construction machinery on the dam surface in the next 6 seconds are 1.17 and 2.36 m,respectively,and the fuzzy-evidence fusion-based potential collision risk grading warning results of the construc-tion machinery provide an automated and intelligent analysis method for the construction safety of the dam surface.
关键词
施工机械安全/大场景视频监控/轨迹预测/模糊-证据融合/分级预警Key words
construction machinery safety/large-scene video surveillance/trajectory prediction/fuzzy-evidence fusion/graded warning引用本文复制引用
基金项目
水利部重大科技专项项目(SKS-2022109)
国家自然科学基金青年科学基金项目(52009089)
出版年
2024
国家科技期刊平台
NSTL
万方数据