摘要
背景模型是交通监控视频中检测运动目标的一种常用方法。混合高斯模型在训练背景模型的过程中效果良好,但其收敛速度较慢。目前各种改进模型,只是提高其初始化的收敛速度;为了加快检测过程中背景改变时的收敛速度,必须实时检测背景是否发生改变,若改变,则需要对模型重新进行初始化。基于以上情况,提出了一种改进的混合高斯模型,该模型不需要重新初始化,避免了实时检测背景是否发生改变的多余步骤,实验结果明显著提高了检测过程中的收敛速度。
Background model is a common method for detecting moving object in traffic surveillance video. The effect of Gaussian Mixture Model used in training background model is good, but its convergence velocity is low. At present, many improved models only accelerate the initial convergent velocity. For accelerating the convergent velocity when background changes in the process of surveillance, the models need to detect whether background has changed or not real time and then to be initialized again if background changes. In this paper we put forward a new improved Gaussian Mixture Model, which needn＇ t be initialized again if background information changes and avoids redundant steps of detecting whether background has changed or not real time. Experiment result of the new model shows the convergent velocity in the process of surveillance is improved evidently.
出处
《中国图象图形学报》
CSCD
北大核心
2008年第11期2139-2143,共5页
Journal of Image and Graphics
基金
国家自然科学基金项目(60273066)
关键词
快速收敛
混合高斯模型
背景模型
目标检测
Fast convergence, gaussian mixture model, background model, object detection
