基于暗原色先验及运动检测的视频去雾方法  

Video haze removal approach based on dark channel prior and object detection

作　　者：刘东辉[1,2] 兰时勇[1,2] 杨红雨[1,2] 吴树霖 朱超军[1,2] 

机构地区：[1]四川大学计算机学院,四川成都610065 [2]四川大学视觉合成图形图像技术国家重点学科实验室,四川成都610065

出　　处：《计算机工程与设计》2014年第1期188-191,197,共5页Computer Engineering and Design

基　　金：国家863高技术研究发展计划基金项目(2012AA011804);四川省科技创新苗子工程基金项目(2012ZZ026)

摘　　要：为了解决视频实时去雾的问题,提出了一种基于暗原色先验及背景差分运动检测的实时视频去雾方法。获得高清视频的背景,采用背景差法获得前景运动目标,利用多线程及GPU加速技术分别求得背景及前景的透射率,将两者的透射率进行融合,根据融合后的透射率还原出无雾视频。实验结果表明,该方法去雾效果明显,对于高清视频(1280*720像素),可以达到20帧/秒以上的处理速度,满足视频实时处理的要求。A real time video dehazing method using dark channel prior and background subtraction is proposed to solve the video real time dehazing problem. The background of the high-definition video is estimated at first, then the moving object is inferred by the background subtraction, the transmittance of the background and the moving object is calculated by multithreading and GPU acceleration technology. The video is recovered based on the fused transmittance of the background and the moving object. Experiment results indicate that this method performs well. Its processing speed is faster than 20 frames/s when dealing with high-definition video of 1280 ~ 720 pixels, which satisfies the demands of real time video applications.

关 键 词：去雾 暗原色先验 背景差分 高清 实时 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]