一种基于全景视觉的运动目标检测改进方法  

An Improved Method of Moving Object Detection Based on Panoramic Vision

作　　者：夏丽娟 姚明磊 张晓玲 XIA Lijuan;YAO Minglei;ZHANG Xiaoling(School of Electrical Information Engineering,Jiangsu University of Technology,Changzhou 213000,China)

机构地区：[1]江苏理工学院机械工程学院,江苏常州213000

出　　处：《电光与控制》2022年第12期47-50,65,共5页Electronics Optics & Control

基　　金：国家自然科学基金(61305123);江苏理工学院研究生实践创新计划项目(XSJCX20_37)。

摘　　要：对视觉领域中全景视觉的不断研究表明,采用鱼眼镜头获取的图像序列所研究的运动目标检测准确率低,受到噪声干扰时鲁棒性不高,针对该问题提出一种基于全景视觉的运动目标检测改进方法。该方法首先采用五帧差分法对图像进行处理,利用相邻5帧之间进行差分来完成前景与背景的分离,有效地减少目标空洞问题;然后在混合高斯模型中提高自适应学习率和更新高斯分布数量,有效克服了检测中出现的重影现象;最后通过形态学处理,得到目标检测结果。实验结果表明,改进的方法提供了比传统的全景图像运动目标检测更可靠的检测结果,目标检测率较高。The continuous research on panoramic vision in the field of vision shows that the accuracy of moving object detection based on image sequence obtained by fisheye lens is low,and the robustness is not high when disturbed by noise.Aiming at the above problems,an improved method of moving object detection based on panoramic vision is proposed.Firstly,the algorithm uses the five-frame difference method to process the image,and uses the difference between adjacent five frames to separate the foreground from the background,thus effectively reducing the target-hole problem.Then,the ghost phenomenon in detection is effectively overcome by improving the adaptive learning rate and updating the number of Gaussian distribution in the Gaussian mixture model.Finally,the target result is obtained by morphological processing.Experiments show that the improved algorithm provides more reliable detection results than the traditional panoramic image moving target detection does,and the target detection rate is higher.

关 键 词：全景视觉 运动目标检测 混合高斯分布 五帧差分 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]