基于概率数据关联的点状动目标CFAR高分辨检测技术研究

High-resolution CFAR Detection of Dim Moving Point Target in IR Image Sequence Using Probabilistic Data Association Filter

本文以多帧检测成功为条件,通过预测目标在下一单帧上可能的存在区域,再在此区域内进行目标的Meyman- Pearson检测。由于信噪比很低,区域内可能会出现很多虚假目标,利用已获知的目标初始信息和数据关联技术,确定其中某一个为真实目标成为关键决策问题。为此,本文将概率数据关联技术应用到图像序列中目标检测领域,得到了重要的理论分析和实验结果。高分辨检测器接收已获知的目标初始信息(目标的初始位置、运动速率以及目标亮度)后,根据概率数据关联技术进行正确测量值的确认,再用卡尔曼滤波器来预测目标在下一帧的可能状态。文中还给出了理论分析、公式推导过程和实真结果。

After successful detection of a target by multi-frame detection scheme, we can predict the possible target area in the next single frame of image, then the NP detection procedure takes place in that area. Due to the low SNR and false alarm rate, maybe too many false targets occur there. There is a critically important decision problem that which of them is true target. To accomplish this task we employ association methods so that achieve the goal of high resolution detection. The transferred information includes the target＇ s initial position, velocities, and luminance. Then the tracker based upon these information uses Kalman filter predict the target＇ s following states. Some theoretical and experimental results are also given in this paper.