节点文献
红外序列图像中运动弱小目标时域检测方法
Small Moving Target Detection in Infrared Image Sequences Based on Temporal Filtering
【作者】 王博;
【导师】 张建奇;
【作者基本信息】 西安电子科技大学 , 光学工程, 2010, 博士
【摘要】 运动弱小目标的检测是成像制导和告警系统的核心技术之一,是提高系统作用距离和检测概率的重要技术手段,探索和研究新的运动弱小目标检测理论以及如何将最新的检测理论应用于运动弱小目标检测仍是一项重要的课题,对现代战争以及未来战争都有重要的军事意义。当目标距离较远时,目标成像较小,可利用的空间分布信息缺乏,检测困难。针对红外序列图像中的运动弱小目标检测问题,本文开展了如下工作:1.分析了红外序列图像中像素点的时域特性。依据红外序列图像中所有像素点的时域变化情况,讨论了目标、晴空背景以及云杂波的时域剖面变化规律,比较了目标、晴空背景以及云杂波均值和方差特性的差别。随后,以傅立叶变换为手段,进一步讨论了不同像素点时域剖面的频谱特性,通过设置不同的带通滤波器对时域剖面进行滤波,研究了晴空背景、云杂波和目标时域剖面在不同频谱范围内的幅频特性。2.提出了一种基于时域剖面滤波的运动弱小目标检测方法。针对时域目标检测算法中跟踪数据量大,实时实现难度高的缺点,给出一种背景移除的方法减少时域检测算法的跟踪数据量。在此基础上,采用时域剖面滤波的方法,去除云杂波时域剖面中较大的起伏。重点分析了下驻点连线(CLSP)滤波法、最小值滤波法、中值滤波法、数学形态学滤波法、均值滤波法、线性滤波法和Savitzky- Golay滤波法等几种时域剖面滤波方法。并进一步分析了时域剖面偏离滤波所得基准的分布特性,得到了一个合适的目标检测量度。最后,还给出了新算法的序贯执行方程。3.研究了空域和时域结合滤波对红外序列图像运动弱小目标检测的改进。首先,介绍了几种典型的基于空域滤波的弱小目标检测方法。其次,将空域目标检测算法中的背景预测算法与提出的基于时域剖面滤波的运动弱小目标检测算法相结合,提出了一种空时域结合滤波的运动弱小目标检测算法。最后,依据红外序列图像中运动弱小目标的运动连续性,构造了一组滤波模板,利用这组模板对检测结果进行滤波,确定出弱小目标可能的运动轨迹。并进一步结合时域特征,实现了运动弱小目标的进一步累积增强。4.探讨了基于动态规划方法的运动弱小目标的能量累积方法。结合现有动态规划算法的优缺点,提出了一种改进的基于动态规划的运动弱小目标检测方法。该方法以目标的运动特性为基础,构造出一个概率模板来描述目标在下一帧可能出现的位置。由于本方法利用概率来描述目标的运动而不是直接硬性的约束,因此,很好地克服了目标运动的随机性。
【Abstract】 Small moving target detection algorithm is one of the most importance key technologies in warning systems and imaging-guidance systems, and also is a significant method to improve the operating range detection probability of the systems. To explore and study the new theory of small target detection, as well as how to test the available theory is an important issue, which has great significance for modern and intending warfare. As a target far away from a detector, the image of the target is small and the information of the spatial is lack, which makes target detecting difficultlly. In this paper, we will investigate the small moving target detection in infrared image sequences. The following works are carried out:1. The characteristic of temporal profile in infrared image pixels are analyzed. Based on the temporal behavior of different types of pixels, the means and the variances of temporal profile are discussed. And of which the differences of clear sky background, cloud clutter, and target are compared. After that, by using the Fourier transform, the spectrum of temporal profile is also analyzed. By setting the different band-pass filter, the temporal profiles of clear background, cloud clutter, and target are filtered, from which we investigate the spectrum characteristic in different frequency bands.2. Several temporal profile based algorithms are proposed. To deal with the drawback of large scale data processing and real-time implementation in temporal filtering, a new background elimination method is presented. Based on this, we proposed to use temporal filtering to remove the impact of the large fluctuation of cloud edges. Detailed analysis is focused on the line of connecting line of the stagnation points based filtering method, the minimum filtering method, median filtering method, mathematical morphology filtering method, average filtering method, linear filtering method and the Savitzky-Golay filtering method. And further, the deviation of the temporal profile and its baseline is analyzed, which lead to a detection criterion. Finally the sequential formulas of the proposed algorithms are also given.3. The combination of spatial and temporal filtering to improve the small moving target detection performance in infrared image sequences is discussed. Several typical spatial filtering algorithms are reviewed. By combining the background prediction based algorithm and temporal filtering based algorithm and considering the continuous of the trajectories of dim point targets a spatial and temporal combined detection algorithm is presented. Based on the analysis of the probable trajectories of moving dim point targets, a group of filter templates are constructed. And the trajectory of dim moving target is obtained by using the constructed template to filter the temporal detection result. Also the target occurrence time in each pixel is extract from the temporal based algorithm to further eliminate interference of background.4. Small moving target accumulated method in image sequences is demonstrated. We analyze the dynamic programming based algorithm for dim signal accumulating in multi-frame image sequences. Considering the advantage and disadvantage of dynamic programming algorithm available, a new small moving target detection algorithm in infrared image sequences is presented to reduce the energy scattering in dynamic programming based algorithm. Based on the property of target motion a Gaussian template is built to model target position in the next frame. Our algorithm uses probability not hard constrain, so it can overcome the randomicity of target motion.
【Key words】 infrared image sequences; small moving target; temporal profile; filter; dynamic programming;