计算机应用 ›› 2015, Vol. 35 ›› Issue (7): 2033-2038.DOI: 10.11772/j.issn.1001-9081.2015.07.2033

• 虚拟现实与数字媒体 • 上一篇    下一篇

基于动态阈值的核密度估计前景检测算法

杨大勇, 杨建华, 卢伟   

  1. 大连理工大学 控制科学与工程学院, 辽宁 大连 116023
  • 收稿日期:2015-01-19 修回日期:2015-03-25 出版日期:2015-07-10 发布日期:2015-07-17
  • 通讯作者: 杨大勇(1988-),男,河北唐山人,硕士研究生,主要研究方向:智能监控、图像处理,dayong657@126.com
  • 作者简介:杨建华(1958-),男,辽宁沈阳人,教授,博士生导师,主要研究方向:城市管网综合自动化系统、智能监控; 卢伟(1976-),男,陕西西安人,讲师,博士,主要研究方向:时间序列分析。
  • 基金资助:

    国家科技重大专项(2011ZX05039)。

Foreground detection algorithm based on dynamic threshold kernel density estimation

YANG Dayong, YANG Jianhua, LU Wei   

  1. School of Control Science and Engineering, Dalian University of Technology, Dalian Liaoning 116023, China
  • Received:2015-01-19 Revised:2015-03-25 Online:2015-07-10 Published:2015-07-17

摘要:

为解决煤层气开采(CBM)现场中抽水机往复运动和风吹草动等动态环境对前景检测的干扰及核密度估计(KDE)目标检测法实时性差的问题,提出了一种改进核密度估计前景检测算法。该方法先用背景差分法(BS)融合三帧差算法将图像分割成动态背景区与非动态背景区,对于动态背景区再用核密度算法分割前景。分割前景时提出了一种新的动态阈值求取方法,综合了相邻样本绝对差均值和样本方差来确定窗宽,并用定时更新与实时更新相结合的策略更新第二背景模型,在替换样本时用随机抽取策略代替先进先出(FIFO)方式。仿真结果表明,改进核密度估计算法与核密度估计法和背景差分核密度估计(BS-KDE)法相比,平均每帧图像算法耗时分别降低了94.18%和15.38%,识别的运动目标也更为完整。实验结果表明所提算法在煤层气开采场景中能准确检测到前景,并基本满足标清视频监控实时性要求。

关键词: 煤层气开采, 往复运动干扰, 核密度估计, 动态阈值, 窗宽, 背景更新

Abstract:

A new improved Kernel Density Estimation (KDE) algorithm used to segment foreground was proposed for the problem of reciprocating pumps and other troubles for segmenting foreground in the field of Coal Bed Methane (CBM) extraction and poor real-time of KDE. Background Subtraction (BS) and three frame difference were applied to divide the image into dynamic and non-dynamic background regions and then KDE was used to segment foreground for the dynamic background region. A new method of determining dynamic threshold was proposed when segmenting foreground region. Mean absolute deviation over the sample and sample variance were combined to compute the bandwidth. And the strategy of combining regular update with real-time update was used to renew the second background model. Random selection strategy instead of First In First Out (FIFO) mode was applied when replacing samples of the second background model. In the simulation experiments, the average time-consuming of per frame image is reduced by 94.18% and 15.38% and moving objects are more complete when comparing the improved KDE with the KDE and Background Subtraction Kernel Density Estimation (BS-KDE) respectively. The experimental results show that the proposed algorithm can detect foreground in the field of CBM extraction accurately and meet the real-time requirement in the standard definition video surveillance system basically.

Key words: Coal Bed Methane (CBM) extraction, reciprocating interference, Kernel Density Estimation (KDE), dynamic threshold, bandwidth, update background

中图分类号: