工厂场景中的异常行为检测  

Abnormal Behavior Detection in Factory Scenarios

作　　者：赵廉 周雷 郭育恒 陈骅桂 ZHAO Lian;ZHOU Lei;GUO Yuheng;CHEN Huagui(School of Health Science and Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China;School of Computer Science and Information Engineering,Shanghai Institute of Technology,Shanghai 201418,China)

机构地区：[1]上海理工大学健康科学与工程学院,上海200093 [2]上海应用技术大学计算机科学与信息工程学院,上海201418

出　　处：《软件导刊》2024年第1期57-62,共6页Software Guide

基　　金：国家自然科学基金项目(61906121)。

摘　　要：针对目前工业场景中的安全生产问题,提出一种异常行为检测框架,主要针对工人睡觉和发生跌倒两种特殊情况。采用人体关键点识别与机器学习分类器相结合的思路,先通过对视频图像中的工人进行关键点识别,提取身体坐标点信息,再训练分类器进行分类,采用多种机器学习方法及一种集成学习模型,实现了对异常情况的检测。在跌倒数据集上,集成学习算法的准确率、精确率和召回率分别达到92.86%、87.58%、98.96%;在睡觉检测方面,算法的准确率、精确率和召回率分别达到98.51%、95.81%、 94.97%。实验表明,该框架能有效检测异常情况,有助于规范生产行为,具有实际应用价值。A framework for abnormal behavior detection is proposed to address safety production issues in current industrial scenarios,mainly targeting two special situations:workers sleeping and falling.The idea of combining human key point recognition with machine learning classi-fiers is adopted.Firstly,key point recognition is performed on workers in video images,body coordinate point information is extracted,and then the classifier is trained for classification.Multiple machine learning methods and an integrated learning model are used to detect abnormal situations.On the fall dataset,the accuracy,accuracy,and recall of the ensemble learning algorithm reached 92.86%,87.58%,and 98.96%,respectively;In terms of sleep detection,the accuracy,accuracy,and recall of the algorithm reached 98.51%,95.81%,and 94.97%,respectively.Experiments have shown that this framework can effectively detect abnormal situations,help standardize production be-havior,and has practical application value.

关 键 词：行为识别 动作检测 异常行为 跌倒检测 机器学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]