基于全景可见光图像的海上弱小目标检测算法研究

In remote place on the sea, targets with large size can only show themselves several pixels in the image, and at this time, the target detection process is called the dim-small targets detection. Panoramic vision system has wide application prospects in maritime surveillance for its advantages of unified imaging, 360-degree view and rotation invariant，but there is lack of corresponding techniques for target detection. Because of the specificity of panoramic visible image and the complexity of sea-sky background, the dim-small surface targets detection based on panoramic vision becomes the difficult problems in the field of target detection and tracking. This paper focuses on the research of the following aspects: A sea-sky line extraction algorithm based on fractal feature and Active Contour Model is launched to solve the problem that the discontinuous oval sea-sky line is difficult to be extracted. The lifting wavelet transform combined with diffusion filter is suggested to detect panoramic dim-small targets around sea-sky line with complexity sea-sky background. Chinese waters is the foggy area in the Pacific Ocean, so dim and small target detection in sea fog in the image can not be avoided. According to the imaging characteristics of panoramic image, Multiple-Scattering Model is established to describe the progress in causing sea fog image and defog sea fog image, so that the dim-small targets in hazy image are detected more efficiently.

在距离很远的情况下,自身尺寸很大的目标在成像平面上也只占很少的像素，此时对目标进行检测的过程就称为弱小目标检测。全景视觉系统以其“成像一体化、360度大视场、旋转不变性”等优点在海域监控领域具有广阔应用前景，但却缺乏相应的目标检测技术支持。全景可见光图像的特殊性和海天背景的复杂性使基于全景视觉的海上弱小目标检测成为目标检测与跟踪领域亟待解决的难点问题，本课题围绕该问题展开研究：针对大范围不连续全景椭圆海天线难以提取的问题，开展基于分形特征和主动轮廓模型的海天线提取算法研究；开展提升小波结合扩散滤波的弱小目标检测算法研究，实现对复杂海天背景下全景海天线区域内弱小目标的高鲁棒性检测；我国海域是太平洋的多雾区，海雾图像中的弱小目标检测不可回避，将根据全景图像成像特点建立描述全景海雾图像生成过程的多散射大气模型，基于该模型进行全景海雾图像复原算法研究，解决全景海雾图像中弱小目标难以检测的问题。

全景视觉系统以其“成像一体化、360度大视场、旋转不变性”等优点在海域监控领域具有广阔应用前景，但却缺乏相应的目标检测技术支持。全景可见光图像的特殊性和海天背景的复杂性使基于全景视觉的海上弱小目标检测成为目标检测与跟踪领域亟待解决的难点问题，本课题围绕该问题展开深入研究：. 首先，设计了基于分形维数的全景设备区提取算法和基于视觉显著图的全景设备区提取算法提取全景海域图像中存在的全景设备区干扰，消除其对海天线检测的不良影响；然后，提出了基于改进主动轮廓模型的提取算法和基于改进Seam Carving的提取算法等海天线提取算法，有效解决大范围不连续全景椭圆海天线难以提取的问题；其次，提出了基于提升小波互能量的检测算法和基于暗通道先验理论的检测算法等小目标检测算法，实现了对复杂海天背景下海天线区域内弱小目标的高鲁棒性检测；再次，建立了大气单散射模型，基于此提出了基于暗通道先验的分区域快速滤波去雾算法和基于大气遮罩估计的改进去雾算法等雾天图像复原算法，实现了对非浓雾情况下的海雾图像的高质量复原；最后，建立了大气多散射模型，提出了一种新的基于大气多散射模型的去雾算法，有效实现了（尤其适用于浓雾情况）海雾图像的高质量复原。. 上述基于全景可见光图像的海上弱小目标检测的相关研究成果，经反复水上试验验证后，可尝试应用于相关水上监控领域。建立的基于多散射大气模型的海雾图像去雾算法，是一种可扩展到其它领域的去雾方法，对解决大雾天气条件下的道路、水上交通监控问题有很好的借鉴性。