基于可变形状基的智慧工地危险行为图像检测算法  

Image Detection Algorithm of Intelligent Construction Site Hazardous Behavior Based on Deformable Basis

作　　者：许瑾璐 黎小英 胡汉敏 XU Jinlu;LI Xiaoying;HU Hanmin(Fuzhou University of International Studies and Trade,Fuzhou 350200,China)

机构地区：[1]福州外语外贸学院,福建福州350200

出　　处：《通化师范学院学报》2025年第2期70-74,共5页Journal of Tonghua Normal University

基　　金：2023年福建中青年教师教育科研项目(JAT231148).

摘　　要：在智慧工地建设中,施工人员危险行为不仅具有空间维度特征,同时也具有特殊的时间维度特征,在特征信息不全的情况下,当前的图像识别方法运用单一形状基或者固定数量的多形状基,难以对非刚体动作的维度特征进行精准判定,进而导致检测准确率下降.鉴于此,该文提出一种基于可变形状基的智慧工地危险行为图像检测算法.首先,利用摄像头采集智慧工地现场图像,运用高斯混合模型区分背景与人体目标,采用可变形状基从动态图像序列中恢复非刚体的三维结构,以此解决维度不关联问题.随后,计算经过平移和旋转运动后的三维场景形变程度,并将人体目标形变程度特征点检测结果输入危险行为检测函数中,从而得到智慧工地危险行为的精准检测结果.实验结果表明:该设计方法的精准度高且误识率低,具备可靠性与应用价值.In the construction of intelligent construction sites,the hazardous behavior and actions of construction personnel not only have spatial dimension characteristics,but also have special temporal dimension characteristics.In the case of incomplete feature information,current image recognition methods use a single shape basis or a fixed number of polymorphic bases,which makes it difficult to accurately determine the dimensional features of non-rigid body actions,leading to a decrease in detection accuracy.In view of this,this article proposes an image detection algorithm of intelligent construction site hazardous behavior based on deformable basis.It solves the problem of dimension independence by using cameras to capture images of intelligent construction sites,using Gaussian mixture models to distinguish between background and human targets,and using deformable bases to restore non-rigid three-dimensional structures from dynamic image sequences.It calculates the degree of deformation of the 3D scene after translational and rotational movements,and inputs the detection results of human target deformation feature points into the dangerous behavior detection function to obtain accurate detection results of hazardous behaviors in intelligent construction sites.The experimental results show that this design method has high accuracy and low misidentification rate,and has reliability and practical value.

关 键 词：高斯混合模型 智慧工地 危险行为检测 目标识别 可变形状基 

分 类 号：TD791[矿业工程—矿井通风与安全]