基于混合信息的运动目标检测优化研究  

Optimization of moving target detection based on mixed information

作　　者：徐武[1] 郭兴 文聪 唐文权 孔玲玲[1] XU Wu;GUO Xing;WEN Cong;TANG Wenquan;KONG Lingling(Institute of electrical and Information Engineering,Yunnan Minzu University,Kuning 650500,China)

机构地区：[1]云南民族大学电气信息工程学院,云南昆明650500

出　　处：《应用科技》2021年第1期66-70,75,共6页Applied Science and Technology

基　　金：国家自然科学基金项目(U1802271)。

摘　　要：运动物体检测过程中,基于颜色信息的运动目标检测受光照突变、阴影以及噪声点大、彩色视频颜色易失真等问题影响。针对以上问题,提出了一种基于混合信息的优化算法。首先,在RGB色彩空间模型基础上采用SOFM(selforganizing feature maps)模型完成彩色视频颜色空间建模,改善颜色空间色彩的真实性;其次,分别利用颜色信息和深度信息为每一个像素设计颜色分类器(color of classifier, CLC)和深度信息分类器(depth information of classifier, CLD),按照对应像素点的混合信息特点再结合上一帧的检测信息以及像素的边缘区域特征,自适应地为分类器分配权重值,达到运动检测的要求效果。本文采用多组视频序列进行实验仿真,实验结果表明:文中改进算法对彩色视频检测过程中存在的问题进行了有效改善,且精度、召回率及F数对比其他算法有明显提高,对视频序列的噪声点有良好抑制作用。In the process of detecting moving objects,the detection of moving objects based on color information is affected by the problems such as light mutation,shadow,large noise point and the color distortion of color video.To solve above problems,an optimization algorithm is proposed in this paper based on mixed information.Firstly,on the basis of RGB color space model,self-organizing feature maps(SOFM)model is used to complete color space modeling of color video,improving the authenticity of color space.Secondly,the CLC classifier and the CLD classifier are designed for each pixel by using color information and depth information respectively.Finally,according to the mixed information characteristics of corresponding pixel points,and combined with the detection information of the previous frame and the edge region characteristics of the pixel,the weighted value is adaptively assigned to the classifier so as to achieve the required result of motion detection.In this paper,multigroup video frame sequences are used for experimental simulation.The experimental results show that the improved algorithm proposed in this paper effectively solves some of the problems existing in the process of color video detection,and the accuracy,recall rate and F number are significantly improved compared with other algorithms.It has a good suppression effect on the noise points of video sequence.

关 键 词：运动目标 混合信息 SOFM模型 分类器 权重 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]