一种改进的基于单高斯模型的红外异常目标检测算法

Improved Infrared Anomaly Target Detection Algorithm Based on Single Gaussian Model

基于单高斯模型的红外异常目标检测算法是一种常见的能自适应更新背景模型的检测算法。该算法对各个像素的输出响应进行高斯建模,通过设定的阈值确定目标像素点是否为前景像素点,从而达到检测的目的。本文在单高斯模型的基础上,提出一种改进的异常检测算法,该算法利用奈曼-皮尔逊准则选取最佳阈值,克服了根据经验值选取阈值的局限性,为最佳判决阈值的选取奠定了理论基础,使得在虚假率一定的情况下,检测概率达到最高。实验证明,将常见的经验阈值与本文确定阈值进行比较,本文算法确定的阈值检测效果更佳。

An infrared anomaly target detection algorithm based on a single Gaussian model is a commonly used detection algorithm that can adaptively update the background model.The algorithm performs Gaussian modeling on the output response of each pixel and determines whether the target pixel is a foreground pixel through a defined threshold to realize detection.This paper proposes an improved anomaly detection algorithm based on a single Gaussian model.The algorithm uses the Neiman-Pearson criterion to define the optimal threshold,which overcomes the limitation of selecting the threshold based on empirical values.The paper lays a theoretical foundation for obtaining the best decision threshold so that under a certain false rate,the detection probability can reach the highest value.Experimental results show that,compared to the commonly experienced thresholds,the threshold determined in this study provides a much better detection effect.