融合视觉特性的交通视频雾霾去除方法研究

Intelligent surveillance system has been widely used in intelligent transportation system, however, fog and haze, especially in recent years, often degrade the video quality, and severely restrict the performance of intelligent surveillance systems, The fog and haze removal for low-quality traffic video has become a problem that calls for immediate solution in intelligent traffic surveillance systems. Therefore, this project will systematically study the fog or haze removal and enhancement method for the low quality videos of traffic surveillance system. Firstly, this project will analyze and build the image degradation model in the daytime and nighttime haze weather, respectively. By analyzing the characteristics and statistical laws of these foggy and hazed images, this project will develop a daytime-haze automatic identification method. Then, from the point of image restoration and enhancement views, this project will develop several image dehazing and wide dynamic range image enhancement methods for daytime and nighttime images by integrating human visual characteristics and maximizing the quality of restored images. Furthermore, a low-quality video dehazing method based on the bilateral temporal-spatial filter will be proposed. Lastly, a new video haze removal and enhancement method will be proposed to improve the low-quality traffic surveillance video.

智能交通视频监控系统已经在智能交通中得到广泛的应用，然而近年频繁发生的雾霾天气导致监控视频质量降低，严重制约和影响了视频监控系统正常有效的工作，交通低质视频雾霾去除成为了智能视频监控系统迫切需要解决的问题。因此，本课题系统地研究面向智能交通监控的低质视频雾霾去除方法。研究建立昼夜不同时刻的大气散射物理模型以及噪声抑制策略；分析雾霾图像特征及统计规律，研究昼间雾霾天气自动识别方法；从图像复原与图像增强两个角度分别研究昼夜不同时段的雾霾去除与噪声抑制方法；为提高雾霾去除质量，融合人眼视觉特性，研究最大化图像增强的雾霾去除与宽动态范围图像增强方法；进而，研究基于时空双边滤波的视频雾霾去除方法，最终形成一套适合交通视频监控系统的雾霾去除与增强新方法。

智能视频监控系统已经在智能交通中得到广泛的应用，然而近年频繁发生的雾霾天气导致监控视频质量降低，严重制约和影响了视频监控系统正常有效的工作，交通低质视频雾霾去除成为了智能视频监控系统迫切需要解决的问题。.本项目系统地分析了雾霾图像退化机理，提出了融合视觉特性的交通低质视频雾霾去除方法，①确定大气传输图的下限界，引入人眼视觉特性，研究了融合视觉对比敏感度的单幅图像雾霾去除与噪声抑制方法；②分析了雾霾图像统计特征，阐明了雾霾图像统计特征与雾霾浓度的内在关系，建立了一种简单的雾霾浓度感知模型，进而提出了一种基于雾霾浓度估计的去雾方法，并分别将其应用到昼夜间图像雾霾去除中；③引入深度学习理论，研究了基于数据驱动的深度去雾方法并将其发展应用于视频去雾中；研究了基于深度置信网络的有效雾霾特征选择与大气传输图估计算法；④提出了低照度条件下图像增强方法；最终建立了一套融合视觉特性的高准确度自适应交通雾霾图像去雾方法。本项目能为交通监控系统提供高效稳定的雾霾去除处理方法，将大幅改善智能视频检测系统有效性与抗噪性；所建立的雾霾去除方法还有望应用于国防视觉监控、遥感图像以及智能成像设备中。