深空背景红外弱小目标检测和跟踪技术研究

发布时间：2018-11-07 12:22

【摘要】：红外成像技术是一种被动接受目标的红外热辐射特性从而获得目标图像的技术,其优点为隐蔽性强、全时段、观测范围大等。随着红外技术的不断发展及深入研究,红外系统已经被广泛应用于监控、制导、跟踪和搜索等军事及民用场景。其中,红外弱小目标检测和跟踪技术是研究热点之一。深空背景下的红外弱小目标具有以下三个特点,因此使得对它的检测和跟踪变得更为困难:?由于成像距离较远,观测到的目标为弱小目标,目标在图像中仅仅占很少的像素数;?在系统噪声和背景噪声的双重干扰下,目标信号一般较弱,并且很容易淹没于强起伏的背景环境中;?由于观测到的是红外弱小目标,目标没有固定的形状,并且缺少纹理信息,大大减少了可提取的目标点特征信息。因此,深空背景条件下的目标检测和跟踪算法研究是一个非常有挑战的课题,对其进行深入研究具有重要的理论意义和实际应用价值。本文紧密围绕深空背景下红外弱小目标检测和跟踪技术进行了深入的研究,主要工作和创新点如下:(1)研究了深空红外图像预处理算法。提出了一种基于方差加权信息熵为复杂度指标的自适应Butterworth高通滤波预处理算法。该算法首先基于方差加权信息熵指标对深空红外图像背景复杂度进行定量衡量,在此基础上提出了深空红外图像背景复杂度的概念,然后利用衡量后的参数,调节Butterworth高通滤波器系数,对简单背景、高噪声点干扰和目标淹没环境下的背景图像进行了自适应处理。实验结果表明,基于方差加权信息熵的Butterworth高通滤波算法可以较好地抑制背景噪声,从而提高图像信噪比。(2)研究了深空背景红外图像弱小目标检测算法。针对传统Otsu算法分割性能不高和实时性差的问题,提出了一种二维直方图斜分Otsu快速迭代算法。该算法首先分析了二维直方图斜分相比于直分算法的优势,然后结合迭代的思想,针对深空背景红外图像目标的特点,利用能量累积算法对其进行优化。实验结果表明,基于能量累积的二维直方图斜分Otsu快速迭代算法能够较好地提升目标检测性能,同时满足实时性要求。(3)研究了深空背景红外图像弱小目标跟踪算法。提出了一种基于多特征融合的优化辅助粒子滤波跟踪算法。该算法针对深空红外弱小目标特征缺乏且难于提取的问题,首先分析了多特征融合的有效性,利用多特征融合的思想,结合不同种类特征的特点来增加目标的信息量;然后采用针对非线性非高斯环境的粒子滤波跟踪算法,通过增加辅助变量的方式优化权值更新,并在算法的滤波过程中嵌入经典的Mean-shift算法,结合多特征融合的优点,从而优化整个跟踪过程。实验结果表明,基于多特征融合的优化辅助粒子滤波跟踪算法,具有较好的准确性和鲁棒性。
[Abstract]:Infrared imaging technology is a kind of passive infrared thermal radiation characteristics of the target to obtain the image technology, its advantages are strong concealment, the entire period of time, a wide range of observations, and so on. With the continuous development and in-depth research of infrared technology, infrared systems have been widely used in surveillance, guidance, tracking and search and other military and civilian scenes. Among them, infrared small and weak target detection and tracking technology is one of the research hotspots. Infrared dim targets in deep space background have the following three characteristics, so it is more difficult to detect and track them:? Because the imaging distance is long, the observed target is a weak target, which accounts for only a small number of pixels in the image;? Under the dual interference of system noise and background noise, the target signal is usually weak and easily submerged in the strong undulating background environment. Because the infrared dim target is observed, the target has no fixed shape, and lacks texture information, which greatly reduces the feature information of the target point that can be extracted. Therefore, the research of target detection and tracking algorithm under the condition of deep space background is a very challenging subject, which has important theoretical significance and practical application value. In this paper, the detection and tracking techniques of infrared dim and small targets in deep space background are deeply studied. The main work and innovations are as follows: (1) the pre-processing algorithm of infrared image in deep space is studied. An adaptive Butterworth high pass filter preprocessing algorithm based on variance weighted information entropy is proposed. The algorithm firstly measures the background complexity of deep space infrared images quantitatively based on variance weighted information entropy index, and then puts forward the concept of background complexity of deep space infrared images, and then uses the measured parameters. By adjusting the coefficients of Butterworth high pass filter, the background images in the environment of simple background, high noise point interference and target inundation are processed adaptively. The experimental results show that the Butterworth high-pass filtering algorithm based on variance weighted information entropy can suppress the background noise and improve the signal-to-noise ratio of the image. (2) the small and weak target detection algorithm of infrared image with deep space background is studied. In order to solve the problem of poor segmentation performance and poor real-time performance of traditional Otsu algorithm, a fast Otsu iterative algorithm with two dimensional histogram diagonal division is proposed. This algorithm firstly analyzes the advantages of two-dimensional histogram oblique division compared with the direct partition algorithm, and then optimizes it by energy accumulation algorithm according to the characteristics of the infrared image of deep space background combined with the idea of iteration. The experimental results show that the fast iterative algorithm of 2-D histogram oblique partition Otsu based on energy accumulation can improve the performance of target detection and meet the real-time requirements. (3) A dim target tracking algorithm for infrared images with deep space background is studied. An optimized particle filter tracking algorithm based on multi-feature fusion is proposed. This algorithm aims at the problem that the feature of infrared dim targets in deep space is scarce and difficult to extract. Firstly, the effectiveness of multi-feature fusion is analyzed. The idea of multi-feature fusion is used to increase the information content of the target by combining the characteristics of different kinds of features. Then the particle filter tracking algorithm aiming at nonlinear non-Gao Si environment is adopted to optimize the weight updating by adding auxiliary variables, and the classical Mean-shift algorithm is embedded in the filtering process of the algorithm, which combines the advantages of multi-feature fusion. In order to optimize the entire tracking process. Experimental results show that the optimized particle filter tracking algorithm based on multi-feature fusion has good accuracy and robustness.
【学位授予单位】：国防科学技术大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.41;TN219

【参考文献】