融合SORM背景模型和DTCNN阈值模型的运动目标检测  

Moving Object Detection Fused with SORM Background Model and DTCNN Threshold Model

作　　者：贾丽娟[1] 

机构地区：[1]西北民族大学数学与计算机科学学院,兰州730030

出　　处：《计算机工程》2016年第1期220-224,共5页Computer Engineering

基　　金：国家自然科学基金资助项目(61163066;60902074)

摘　　要：为降低动态背景对运动目标检测性能的影响,提出一种运动目标检测方法。该方法融合自组织视网膜映射图(SORM)背景模型和离散时间卷积神经网络(DTCNN)阈值模型。依据SORM构建背景模型,结合运动目标检测要求,改进DTCNN算法,构建阈值模型,在检测运动目标的过程中降低噪声干扰,自适应更新SORM和DTCNN模型中的相关参数,以适应场景变化。实验结果表明,与经典的高斯混合模型、自组织背景差分和增长自组织映射图方法相比,该方法的目标检测性能更好,尤其是对动态背景的适应能力更强。For reducing the impact on the performance of moving object detection by dynamic background,a moving object detection method is proposed.This method is fused with Self-organizing Retinotopic Maps（SORM） background model and Discrete Time Cellular Neural Network（DTCNN） thresholding model,builds background model according to SORM,ameliorates DTCNN in terms of requirement of moving object detection,constructs thresholding model to remove noise during the process of moving object detection,and updates the parameters of SORM and DTCNN models to adapt scene change.Experimental results show that this method has better performance of object detection,especially has stronger adaptation capacity for dynamic background,compared with traditional methods such as Gaussian mixture model,self-organizing background subtraction and growing self-organizing maps.

关 键 词：目标检测 背景模型 阈值模型 自组织视网膜映射图 神经网络 动态背景 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]