保护电梯门的动态图像识别方法  

A New Image Identification Method for Elevator Door Protection

作　　者：吕新知 谷明非[1] 宋世杰 宫大为 LV Xin-zhi;GU Ming-fei;SONG Shi-jie;GONG Da-wei(Nuclear Power Institute of China,Chengdu 610213,China;School of Mechanical and Electrical Engineering,University of Electronic Science and Technology of China,Chengdu 611731,China)

机构地区：[1]中国核动力研究设计院,四川成都610213 [2]电子科技大学机械与电气工程学院,四川成都611731

出　　处：《软件导刊》2021年第5期158-163,共6页Software Guide

基　　金：武器装备预研基金项目(1126190402A,1126170104A,1126180204B,1126190508A)。

摘　　要：针对用于电梯门保护的图像处理技术存在信息量大、动态特性强等问题,提出一种适用于电梯门保护系统的动态图像识别方法。改进了中值滤波方法,有效过滤了图像噪声并可保持边界特征;采用类间方差法对图像进行全局阈值分割,建立以背景差分法为主、相邻帧差法为辅的动态目标检测模型,采用区域过度生长算法实现动态目标跟踪和矢量方向判断。建立电梯门保护系统试验平台并进行试验验证。结果表明,该方法可在3帧图像内完成目标识别,对目标慢速、中速、快速移动判别的正确率分别达到83.3%、93.3%和100%。Oriented to large storage capacity and strong dynamic characteristics of image technology in the process of elevator door protection,a new image identification method is presented brought forward.Modified median filtering method is established to filter the image noise and to hold on the boundary clearly.Otsu method is adopted for image threshold segmentation.Relying on background subtraction method and temporal difference method,the dynamic target identification model is set up.Combined with region growing segmentation algorithm,the dynamic target’s movement is detected on time.The test platform is built up for the elevator door protection system.The experimental results show that the image identified process could be finished by detecting 3 frames of images and the accuracy of discrimination for slow,medium and fast moving targets could reach 83.3%,93.3%and 100%,respectively.

关 键 词：图像识别技术 改进的中值滤波方法 类间方差法 

分 类 号：TP317.4[自动化与计算机技术—计算机软件与理论]