基于改进粒子滤波的煤矿视频监控模糊目标检测

发布时间：2018-03-01 02:36

本文关键词： 信息处理技术目标检测粒子滤波多状态空间关键区域采样　出处：《吉林大学学报(工学版)》2017年06期 　论文类型：期刊论文

【摘要】：针对煤矿井下智能视频监控采集到的视频图像质量较差、干扰多、噪点多、目标检测不准确等问题,提出了一种基于改进粒子滤波的模糊目标检测方法。在标准粒子滤波理论框架下,以视频多图像帧差为基础构建非线性、非高斯多系统状态空间融合模型;在图像帧差得到的关键点区域范围内进行粒子抽样及概率密度的传播;利用加权后验样本粒子表示多系统状态空间融合模型的后验概率密度;采用样本均值方法融合估计系统后验状态;最后对系统状态空间模型进行输出,得到目标检测结果。使用三交河煤矿井下视频监控数据进行试验验证,分别懫用改进的粒子滤波、标准粒子滤波、无损卡尔曼滤波以及扩展卡尔曼滤波等方法进行对比。试验结果表明:针对煤矿井下视频模糊目标检测问题,改进的多状态空间模型及关键区域采样粒子滤波算法具有良好的目标检测效果。
[Abstract]:In the coal mine intelligent video surveillance capture video images of poor quality, interference, noise, target detection is not accurate, this paper proposes a fuzzy target detection method based on improved particle filter. In the framework of the standard particle filter, the multi video like frame difference to construct nonlinear based multi system state space the fusion model of non Gauss; in the image frame difference area key points that are spread within the particle sampling and probability density; using weighted posterior probability density test sample particle representation model of state space fusion multi system; using the method of sample mean posterior state fusion estimation system; finally, the output of the system state space model and get the target detection results. Experiment is carried out to verify the use of video surveillance data of three underground coal mine, Zhi respectively with improved particle filter, standard particle filter, no Compared with Calman filtering and extended Calman filtering, experimental results show that the improved multi state space model and key area sampling particle filter algorithm have good target detection effect for underground mine video fuzzy target detection problem.

【作者单位】：中国矿业大学(北京)机电与信息工程学院;安徽理工大学经济与管理学院;
【基金】：国家自然科学基金项目(51404007,61601466) 安徽省重大教学改革研究项目(2015zdjy074)
【分类号】：TN948.6;TP391.41

【相似文献】