无人机监测下厂区安防入侵识别  

Identification of security intrusion in factory area under drone monitoring

作　　者：陆正卿 方维岚 顾凯 LU Zhengqing;FANG Weilan;GU kai(Shanghai Tobacco Group Co.,Ltd.,Shanghai 200082,China)

机构地区：[1]上海烟草集团有限责任公司,上海200082

出　　处：《电子设计工程》2025年第7期186-191,共6页Electronic Design Engineering

摘　　要：当前厂区周界安防管理未考虑目标停留点特征差异,使得识别结果的F1分数(F1 Score)较低。为了提升识别结果的F1分数,保证安防效果,提出基于无人机数据与语义轨迹挖掘的厂区周界安防入侵事件识别方法。通过采集无人机监测数据并处理,采用改进背景差分法检测图像中包含的安防入侵目标,结合像素分类原理获取入侵目标运动轨迹跟踪结果。将大量运动跟踪轨迹转换为语义轨迹,在考虑停留点兴趣度、目标停留时间、目标语义轨迹支持度的前提下,应用蚁群算法搭建语义轨迹挖掘模式,识别出厂区周界安防入侵事件。实验结果表明,所提方法识别结果的F1分数高于0.9,基本满足了厂区安防检测要求。The current perimeter security management of the factory area has not taken into account the differences in target stopping point characteristics,resulting in a low F1 Score in the recognition results.In order to improve the F1 Score of recognition results and ensure security effectiveness,a method for identifying perimeter security intrusion events in factory areas based on drone data and semantic trajectory mining is proposed.By collecting drone monitoring data and processing it.Using an improved background difference method to detect security intrusion targets contained in the image,and combining pixel classification principles to obtain intrusion target motion trajectory tracking results.Convert a large number of motion tracking trajectories into semantic trajectories,and apply ant colony algorithm to build a semantic trajectory mining mode,taking into account the interest of dwell points,target dwell time,and support for target semantic trajectories,to identify perimeter security intrusion events in the factory area.The experimental results show that the F1 Score recognized by the proposed method is higher than 0.9,which basically meets the requirements of factory security detection.

关 键 词：无人机数据 语义轨迹挖掘 厂区 安防系统 入侵事件 识别方法 

分 类 号：TN964.1[电子电信—信号与信息处理]