基于大气微点光源主动照明退化模型的斜视航拍宽幅雾霾场景清晰化方法研究

The widely existence of haze weather leads to the Oblique aerial Large image degradation, which greatly limits the effectiveness of Unmanned Aerial Vehicle (UAV) Imaging system. Therefore, research on haze scene recovery method has important practical significance. In the haze imaging path with squint aerials, atmospheric non-uniform and multi-scattering effects cannot be ignored, the microscopic particles which participating in multiple scattering in the atmosphere are regarded as numerous active illumination micro point sources. A new non-uniform and multi-scattering model is established by convolving these micro point source energy with the atmospheric point spread function and superimposing them on the imaging field of view; the image layering method is used to separate the Gaussian low-pass filter estimated halo image from the original image; the atmospheric light images are estimated based on the characteristics of the bright channel and refined by guided filtering; using dark channel priors rough estimate transmittance image, and using the multi-scale edge-aware filter which based on Bayesian model averaging method to get the fine estimation of transmission map; A similar logarithmic transformation of the generalized computing model in the bounded range is used to enhance the contrast of dehazing image; complete algorithm parallel transformation on GPU platform, achieve rapid recovery of large haze scene. This project study on the haze scene fast recovery method that can enhance the situational awareness capabilities and improve the robustness and reliability of aerial imaging system in haze weather.

雾霾天气广泛存在，致斜视航拍宽幅图像严重降质，极大地限制了无人机航拍系统效用的发挥。因此，对雾霾场景快速清晰化方法展开研究具有重要现实意义。在斜视航拍宽幅雾霾成像路径中，大气非均匀性与多重散射作用不可忽略，本项目拟将大气中参与多重散射的微小粒子看成无数个主动照明的微点光源，通过将这些微点光源能量与大气点扩散函数卷积并叠加于成像视场中，建立一种新的基于大气微点光源主动照明退化模型；基于该模型采用图像分层方法剔除高斯低通滤波器估计的光晕图像；利用亮通道估计大气光图像并采用引导滤波进行细化；利用基于Bayesian模型平均法的多尺度边缘感知滤波器对暗通道估计的透过率图像进行精细估计；在有界范围利用广义类对数变换得到动态范围大且细节清晰的去雾图像；完成算法在GPU平台上的并行优化，实现宽幅雾霾场景的快速复原。本项目的研究能增强无人机航拍系统的态势感知能力，提高其在雾霾天气下工作的鲁棒性和可靠性。

随着遥感技术的发展，遥感图像已得到广泛应用，然而可见光遥感图像的质量容易受到霾、雾、烟、云等不利大气条件的影响。即使在晴天，大气也不可能绝对纯净，由于大气的散射作用将会导致遥感图像质量下降，影响其效用的发挥，因此遥感图像去雾是一项至关重要且不可或缺的预处理任务。本项目研究了斜视航拍宽幅遥感图像的退化机理，将大气多重散射作用看作是大气中存在无数个主动照明的微点光源，大气的非均匀性导致了这些微点光源的密度和照明强度在空间上分布不同，建立了一种基于大气微点光源主动照明的退化模型；研究了基于分段模型变换的图像去雾算法，通过对暗区建立二次函数模型，对亮区建立线性变换模型，从而得到无雾图像的最小值通道图像，基于有雾图像以及无雾图像的最小值通道图像来估计大气透过率以及大气光,能快速恢复无雾图像；研究了基于不同颜色空间的图像快速去雾算法，根据HSV空间和RGB彩色空间的特性实现有雾场景的复原；研究了基于均小值滤波的图像去雾算法，针对全色遥感图像，采用一种均小值滤波方法，得到透过率的准确估计，最终获得边缘细节清晰、信噪比高的清晰化遥感图像；研究了一种基于低秩-稀疏先验的图像去雾算法，将有雾图像暗通道分解为稀疏的直接衰减暗通道和低秩的大气光幂两部分，通过低秩-稀疏的先验知识复原无雾图像；研究了GPU以及FPGA平台上的并行优化与实现，有效地提高了图像去雾算法的处理效率，为在实际工程中的应用提供了保障。