基于视频的空中目标检测技术研究

【作者】 李磊；

【导师】 赵高鹏；

【作者基本信息】 南京理工大学 ， 控制工程（专业学位）， 2019， 硕士

【摘要】 近年来,视频空中目标检测技术作为防空作战光电火控系统中的核心技术之一,在精确制导、光电搜跟等领域有着广泛的应用。本文以某项目背景为依托,展开基于视频的空中目标检测技术研究,主要包括低空飞行目标检测、目标跟踪以及空中弱小目标检测。针对多云、接地等复杂天空场景下的低空飞行目标,提出了一种基于梯度和局部对比度的空中目标定位方法,依据空中目标与云层边缘、接地背景在局部图像结构上的区别,在梯度计算的基础上,加入了局部对比度,仿真结果表明该方法可以有效抑制云层边缘、接地背景;在目标识别环节,设计了一种基于支持向量机的空中目标识别方法,提取方向梯度直方图特征,采用支持向量机对空中目标区域进行分类,可以有效识别飞机、飞鸟、无人机、导弹四类典型空中目标。针对快速运动、遮挡等跟踪难点,设计了一种基于核相关滤波的长时间跟踪方法。在核相关滤波快速跟踪方法的基础上,加入了检测模块,依据跟踪置信度实时决定是否重新检测,实现了对空中目标的长时间稳定跟踪。针对空中弱小目标存在成像面积小、缺乏纹理信息、边缘干扰等问题,提出了一种基于结构张量和高斯差分的小目标检测方法。依据边缘、小目标以及平滑背景在局部图像结构上的区别,计算出结构张量图;采用高斯差分带通滤波方法计算出高斯差分图;通过归一化点乘得到响应图。仿真结果表明该方法可以有效抑制背景、增强目标。在上述方法研究的基础上,编写了视频空中目标检测的软件,并将其移植到AM5728平台,完成了软件调试和测试,形成了一套较为完整的嵌入式图像处理平台,实验测试验证了该图像处理平台的有效性和可靠性。更多还原

【Abstract】 In recent years,video aerial target detection technology,as one of the key technology of Photoelectric Air Defense Fire Control System,has been widely used in many areas such as precision guidance,photoelectric search and tracking.Based on the background of a certain project,this paper focuses on the research of video aerial target detection technology,including low-altitude flying target detection,target tracking and aerial small target detection.Aiming at the low-altitude flying target in complex sky scenes such as cloudy and grounding,an aerial target localization method based on gradient and local contrast is proposed.According to the difference of local image structure between aerial target,cloud edge and grounding background,local contrast is added on the basis of gradient calculation.The simulation results show that the method can effectively suppress the cloud edge and grounding background.In the aspect of target recognition,an aerial target recognition method based on support vector machine is designed.Histogram of oriented gradient is extracted,and the aerial target area is classified by the support vector machine.It can effectively recognize four kinds of typical aerial targets: aircraft,flying birds,UAV and missile.Aiming at the problems of tracking such as fast motion and occlusion,a long-term tracking method based on kernelized correlation filters is designed.Based on high-speed tracking with kernelized correlation filters,a detection module is added to determine whether to re-detect according to the tracking confidence in real time,and the long-term stable tracking of aerial target is realized.Aiming at the problems of aerial small target such as small image area,lack of texture information and edge interference,a small target detection method based on structure tensor and Gaussian difference is proposed.According to the difference of local image structure between edge,small target and smooth background,the structure tensor map is calculated.The Gaussian difference map is computed by adopting difference of Gaussians band-pass filter.The response map is obtained by normalizing and fusing the two maps.The simulation results demonstrate that the method can effectively suppress the background and enhance the target.Based on the above methods,the software for video aerial target detection is compiled and transplanted to the AM5728 platform.The software debugging and testing are completed,and a relatively complete embedded image platform is formed.The experimental results verify the effectiveness and reliability of the image processor.更多还原