【作者】 李俊；

【导师】 颜露新；

【作者基本信息】 华中科技大学 ， 导航制导与控制， 2017， 硕士

【摘要】 动平台红外成像具有隐蔽性高、抗干扰性好、全天时全天候工作等特点,被广泛应用于侦察、监视等军事领域。动平台下红外成像运动目标检测,作为红外目标识别、跟踪、行为分析、制导等后续处理的基础,一直受到广泛关注。本文针对“平台动、目标动、背景动”条件下不同背景、不同目标特性的红外运动目标检测展开研究,包括中小型红外运动目标检测和远距离弱小红外运动目标检测。主要研究内容包括以下四个方面:首先,根据动平台下红外图像成像特性,对运动目标检测做预处理工作。针对红外成像质量较差时运动目标检测易受噪声干扰的问题,对多种图像去噪算法效果进行研究。针对动平台成像时运动检测需进行帧间图像序列配准问题,探讨了不同特征算子对图像配准的影响。同时,就运动目标检测结果定量分析提出了相关评价指标。其次,结合传统的背景建模方法对红外图像进行运动目标检测,分别利用LBP背景建模和低秩表达背景建模构建了动平台下运动目标检测方法。该类方法适用于非实时性检测,需要将序列图像通过图像配准组成小序列,进行分批次建模处理,可以取得较准确的处理结果,但计算复杂度高,难以做到实时检测。然后,为了适应红外运动目标检测的实时性要求,根据红外图像特点,改进了帧差分检测。通过分析红外序列图像帧间的非均匀性亮度变化特点,在帧间配准后进行亮度线性回归校正,提高差分的准确性。并结合梯度统计直方图对差分结果进行校正,剔除大部分虚警,提高了检测精度。最后,针对远距离红外弱小运动目标,基于单帧图像进行检测,提出利用局部亮度差和局部加权熵,结合多尺度信息进行背景抑制的弱小目标检测方法。实验结果表明,该方法能有效增强目标并抑制背景杂波,提高目标信杂比和目标局部对比度。更多还原

【Abstract】 Infrared imaging from mobile platform with high concealment,all-weather all-day work and other characteristics,is widely used in reconnaissance,surveillance and other military fields.Infrared moving target detection,as the basic work of infrared target recognition,tracking,behavior analysis,guidance,has been widespread concern.This paper aims at the detection of infrared moving objects with different background and different target characteristics under the condition of "moving platform,moving target and moving background".The main research contents include the following four aspects:First,preprocessing work on moving target detection is discussed.Aiming at the problem that the moving target detection is susceptible to noise,the effects of several image denoising algorithms are compared.In order to solve the problem of image registration,the influence of several different feature operators on image registration is discussed.The quantitative evaluation of moving target detection is put forward.Second,based on the traditional background modeling method,LBP background modeling and Low-Rank representation background modeling is used for moving target detection under motion imaging platform.This method is suitable for non-real-time detection,can be achieved accurate detaction results,but the computational complexity is higher.Third,the moving target detection with frame difference is improved based on infrared image characteristics.By analyzing the nonuniformity luminance characteristics of infrared sequence images,we use linear regression to fit the relationship between infrared sequence images.And,the gradient statistics histogram is used to correct the result of moving target detection with frame difference,improved the detection accuracy.Last,for single-frame moving infrared small-target detection,we propose an effective small-target detection approach based on multiscale gray difference and weighted image entropy.The experimental results show that the proposed method can effectively enhance the target and suppress the background.更多还原