论文部分内容阅读
针对复杂背景红外图像序列目标检测的难题,给出了一种用于红外监控系统中入侵目标检测的背景建模方法。应用特征样本集为每一个像素建立统计无参数样本集模型,根据核函数估计计算每一个像素值对模型的符合概率。使用双阈值进行目标检测和模型更新,将图像分为三类:可靠背景、感兴趣区域和不可靠背景。通过不可靠背景类提供的信息进一步将感兴趣区域细分为入侵目标和错误检测。对几种红外图像序列仿真实验表明,该算法不仅可以比较精确的检测显著入侵目标,对于容易淹没在噪声中的弱小入侵目标也可以实现准确地检测。
Aiming at the problem of target detection in complex background infrared image sequences, a background modeling method for intrusion detection in infrared surveillance system is presented. The feature sample set is used to establish a statistical parameter-free sample set model for each pixel. The coincidence probability of each pixel value with the model is calculated according to the kernel function estimation. Target detection and model updating using double thresholds classify images into three categories: Reliable Background, Regions of Interest, and Unreliable Background. The information provided by the unreliable background class further subdivides the area of interest into intrusion targets and error detection. Experiments on several infrared image sequences show that the proposed algorithm not only can detect significant intrusion targets more accurately, but also can accurately detect weak intrusion targets that are easily submerged in noise.