针对现有周界入侵检测技术应用于私人住宅时面临的几个问题,提出一种面向庭院式住宅的周界入侵综合识别算法,该算法以YOLOv5为基础,将周界入侵检测任务分为异常目标检测与入侵检测两部分.在异常目标检测部分,针对YOLOv5在复杂户外场景下对小目标检测效果不佳等问题对其网络结构进行三点改进,然后应用改进YOLOv5对监控范围内人员穿戴与所持工具是否异常进行检测,并根据结果判断是否存在潜在入侵行为;在入侵检测部分,提出一种点线式区域入侵检测方法,巧妙地将区域入侵检测问题抽象成点与多条直线围成区域的位置关系判断问题.实验结果表明:本文算法在异常目标检测阶段平均精度(mean average precision,mAP)为85.4％,相较于YOLOv5与其他目标检测算法精度更高;模型检测速度可达23.4帧/s,实时性良好;在入侵检测阶段,点线式区域入侵检测方法相较于现有基于视频的入侵检测方法具有更高的入侵检测灵敏度且不存在误检现象;基本满足对庭院式住宅进行周界入侵检测的任务需求.
Integrated Perimeter Intrusion Identification Technology for Courtyard Houses
Aiming at several problems faced by the existing perimeter intrusion detection technology when applied to private houses,an integrated detection algorithm for perimeter intrusion for courtyard houses was proposed.The algorithm was based on YOLOv5,and the perimeter intrusion detection task was divided into two parts:abnormal object detection and intrusion detection.In the abnormal ob-ject detection part,the network structure of YOLOv5 was improved by three points in view of its poor detection effect on small objects in complex outdoor scenes.Then,the improved YOLOv5 was applied to detect whether the tools worn and held by personnel within the monitoring range are abnormal,and the potential intrusion behavior was determined according to the results.In the part of intrusion de-tection,a point-line area intrusion detection method was proposed,which cleverly abstracts the problem of area intrusion detection into the judgment of the position relationship between a point and the area enclosed by multiple lines.The experimental results show that the mean average precision(mAP)of the proposed algorithm in the abnormal object detection stage is 85.4％,which is higher than that of YOLOv5 and other object detection algorithms.The model checking speed can reach 23.4 frame/s,and the real-time performance is good.In the intrusion detection stage,compared with the existing video-based intrusion detection methods,the pointline area intrusion detection method has higher intrusion detection sensitivity and no false detection phenomenon.It basically meets the task requirements of perimeter intrusion detection for courtyard houses.
万方数据
维普
