复杂背景下基于LBP纹理特征的运动目标快速检测算法
投稿时间:2021-07-23  修订日期:2021-09-02  点此下载全文
引用本文:
摘要点击次数: 73
全文下载次数: 0
作者单位邮编
裘莉娅 中国科学院上海技术物理研究所 200083
陈玮琳 中国科学院上海技术物理研究所 
李范鸣 中国科学院上海技术物理研究所 200083
刘士建 中国科学院上海技术物理研究所 
李争 中国科学院上海技术物理研究所 
谭畅 中国科学院上海技术物理研究所 
基金项目:中国科学院青年创新促进会资助
中文摘要:在雨雪天气、树叶晃动、水面闪烁等有复杂背景的可见光与红外场景中,快速准确地提取完整目标一直是运动目标检测中的首要难题。为了满足实时性,并针对现有视频的前景提取算法依赖先验信息、召回率低、缺乏纹理和噪声较大等问题,提出了一种基于直方图统计和改进的LBP(Local Binary Pattern)纹理特征相结合的背景建模方法。首先,使用各像素直方图的众数作为参考背景无需先验知识,节省了大量存储空间,再采用邻域补偿策略提出了一种改进的S_MBLBP纹理直方图与参考背景进行背景建模,消除了大部分动态背景和光照变化影响,实现目标的精确提取。实验表明,所提的算法在红外和可见光的多种复杂场景下,能快速提取前景目标的同时,提高了的准确率和召回率。
中文关键词:机器视觉  背景建模  LBP纹理特征  运动目标检测  复杂背景
 
Fast moving target detection algorithm based on LBP texture feature in complex background
Abstract:In the visible and infrared scenes with complex background, such as rain and snow weather, leaf swaying, shimmering water, etc., fast and accurate extraction of a complete target has always been the primary problem in moving target detection. In order to be real time and aiming at the problems of existing video foreground extraction algorithms, such as dependence on prior information, low recall rate, lack of texture and large noise, a background modeling method based on histogram statistics and improved LBP(Local Binary Pattern) texture features is proposed. Firstly, the mode of each pixel histogram is used as the reference background without prior knowledge, which saves a lot of storage space. Then, an improved S_MBLBP texture histogram is proposed to model the background with the reference background by using neighborhood compensation strategy, which eliminates most of the dynamic background and illumination changes, and realizes the accurate extraction of the target. Experimental results show that the proposed algorithm can quickly extract foreground targets in a variety of complex infrared and visible scenes, and can improve the accuracy and recall rate at the same time.
keywords:machine vision,background modeling, LBP textural features, moving target detection, complex background
  HTML  查看/发表评论  下载PDF阅读器

《红外与毫米波学报》编辑部