改进视觉背景提取模型的运动目标检测算法  

Improved Visual Background Extractor Model for Moving Objects Detecting Algorithm

作　　者：苏延召[1] 李艾华[1] 姜柯[1] 金广智[1] 

机构地区：[1]第二炮兵工程大学502教研室,西安710025

出　　处：《计算机辅助设计与图形学学报》2014年第2期232-240,共9页Journal of Computer-Aided Design & Computer Graphics

摘　　要：针对视觉背景提取模型对动态场景适应性不强、运动目标检测精度低等问题,提出一种改进的视觉背景模型算法.在模型建立与初始化阶段,采用按序抽取的方法将像素点本身信息加入到背景模型中,形成邻域背景模型,降低复杂场景对模型的影响;在前景检测阶段,结合像素点的空间邻域信息自适应地获取分割阈值,减少各类复杂场景对检测结果的干扰,提高运动目标检测的精度;在背景更新阶段,根据场景复杂度动态地调整模型的更新周期与更新方式,使得模型能够有效地消除虚影与背景噪声的影响,增强模型对复杂场景的适应性与鲁棒性.与典型算法进行对比的实验结果表明,该算法具备较高的检测精度,适用于动态场景中的运动目标检测.As visual background extraction model is not adaptable enough for dynamic scenes and provides with low accuracy in moving target detection, an improved model of the visual background algorithm is proposed in this paper. In model initialization, neighborhood background model is established by adding the sequentially extracted pixel information to bring down the impact of complex scenes. In order to reduce the interference of complicated scenario on the test results and improve the accuracy of motion target detection, the segmentation threshold is adaptively obtained by combined with the spatial neighborhood information. To enhance the robustness and adaptability in complex scenes, the update cycle and update method of the model are dynamically adjusted according to the scene complexity in the process of background update. The influence of ghost and background noise can be effectively eliminated by updating. Experimental results show that the improved algorithm is applicable to moving object detection in dynamic scenarios and provides with higher detection accuracy.

关 键 词：视觉背景提取 目标检测 自适应阈值 动态更新 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]