面向实时视频去雾处理的关键技术研究

在雾霾天气下，由于浑浊媒介的影响，常规摄像机所采集的视频通常对比度偏低，这将导致机器视觉系统的执行精度急剧下降。因此，使有雾视频还原为清晰场景具有十分重要的现实意义。然而，现有视频去雾框架所采用的成像参数估计策略不但具有极高的计算复杂度，而且其不精确的估计参数将导致恢复视频中出现诸如亮度闪烁、过饱和及过增强等负面视觉效应。为此，本课题以时空相关性为切入点，针对上述问题对视频去雾框架中的参数估计环节展开研究，提出一种以图像组为基本处理单元的成像参数估计理论与方法。主要研究内容包括：入射光数学模型的构建、关键帧透射率的全局估计以及非关键帧透射率的映射机制。课题目标是通过改善去雾质量与提高处理效率，以提供满足高质量用户体验的视频服务。同时，本课题的开展将为智慧城市，卫星遥感，野外侦察，智能驾驶等应用提供可靠的技术支撑。

Under hazy weather conditions, due to the influences of atmospheric suspended particles, videos captured by traditional camera are often plagued by low visibility, which is bound to seriously degenerate the performance of computer vision applications. Therefore, it is of great practical significance to restore blurred hazy video to clear scenes. However, the imaging parameter estimation strategy adopted by the existing video defogging framework not only has extremely high computational complexity, but its inaccurate estimation parameters will lead to negative visual effects such as brightness flicker, over-saturated and over-enhancement into the recovered video. To this end, based on the space-time correlation feature, this project studies the key technical part of the parameter estimation involved in the dehazing frame, and then proposes an image-group-based imaging parameter estimation theory and method. The topics we plan to explore include the establishment of variable incident light model, global estimation of key frame transmission, and mapping mechanism of non-key frames transmission. The goal of the project is to provide a video service that meets a high-quality user experience by improving the quality of dehazing and increasing processing efficiency. Meanwhile, this project will provide the reliable technical support for the actual applications such as smart cities, satellite remote sensing, field reconnaissance, and intelligent driving.