计算机工程与应用 ›› 2017, Vol. 53 ›› Issue (21): 179-184.DOI: 10.3778/j.issn.1002-8331.1605-0209

• 图形图像处理 • 上一篇    下一篇

基于背景分类的运动目标检测算法

高红红1,曹建荣1,2,李振宇1,杨红娟1,2,赵淑胜1   

  1. 1.山东建筑大学 信息与电气工程学院,济南 250101
    2.山东省智能建筑技术重点实验室,济南 250101
  • 出版日期:2017-11-01 发布日期:2017-11-15

Moving target detection algorithm based on background classification

GAO Honghong1, CAO Jianrong1,2, LI Zhenyu1, YANG Hongjuan1,2, ZHAO Shusheng1   

  1. 1.College of Information and Electrical Engineering, Shandong Jianzhu University, Jinan 250101, China
    2.Shandong Provincial Key Laboratory of Intelligent Building Technology, Jinan 250101, China
  • Online:2017-11-01 Published:2017-11-15

摘要: 针对光线暗、对比度和分辨率低的监控视频,提出了一种基于背景分类的运动目标检测算法。 首先用视频第一帧图像HSV空间的色度H和亮度V作为背景特征进行初始化,建立两种包含色度和亮度特征的背景模型类,即初始化得到的原始背景类和受光照或者其他因素影响得到的在原始背景周围波动的背景波动类,利用这两个背景模型进行前景检测和背景更新。为提高前景检测的准确率,背景模型的更正加入背景更正机制和权重机制,使得背景中样本的数量根据背景的实际情况处在一种动态的变化中,提高前景分割的效率。用不同场景下的监控视频进行算法对比实验,结果证明,该算法获得的前景完整清晰,视频处理的速度较快。提出的算法简单实用,对噪声干扰表现出良好的鲁棒性。

关键词: 背景分类, 背景更正机制, 权重机制, 运动目标检测

Abstract: Aiming at surveillance videos which have low light, low contrast and low resolution, this paper proposes a moving target detection algorithm based on background classification. First, the video image hue H and brightness V of HSV in the first frames are used to initialize the background characteristics in order to build two kinds of background model classes including hue and brightness characteristics, namely the original background class that is obtained in the initialization background and the background fluctuating class influenced by lighting or other factors around the original background. These two models are used to detect foreground and to update background model. To improve the accuracy of the foreground detection, the background correction mechanism and weighting mechanism are utilized to correct the background model, so the number of samples in the background changes according to the actual situation in the background in order to improve the efficiency of the foreground segmentation. Experimental results show that the algorithm can obtain complete and clear foreground image and fast processing speed in different scenarios. This algorithm is simple and practical, and has better robustness for noise.

Key words: background classification, background correction mechanism, weight mechanism, moving target detection