【作者】 李娜；

【导师】 钱步仁； 曹旭东； 刘永梅；

【作者基本信息】 中国石油大学(北京) ， 新一代电子信息技术（专业学位）， 2023， 硕士

【摘要】 随着计算机视觉系统的不断发展,数字图像视频广泛应用于我们的日常生活中。然而在雾霾等恶劣天气的影响下,拍摄图像和视频时,雾霾中的大气颗粒会对大气光产生吸收和散射作用,使得采集到的图像视频出现严重的降质问题,造成图像信息模糊和对比度下降等现象,进而对例如交通监控等日常生活产生影响。因此需要对此类图像视频进行去雾处理,复原出清晰的图像。本文将通过对暗通道先验算法进行改进,结合硬件FPGA(Field Programmable Gate Array)对图像视频进行去雾,利用FPGA的并行计算特性加速图像视频去雾。本论文首先对国内外的去雾处理算法及其在FPGA上实现的相关研究领域、研究现状等进行了调研与分析,剖析了整个图像去雾领域的各种研究算法与平台,得出了有雾天气情况下降质图像视频的退化原因。其次建立物理模型,提出一种改进的分窗加权暗通道先验自适应阈值去雾算法,加权融合初始暗通道值与最小值滤波的暗通道值,设置阈值求取大气光,引入修正因子修正透射率的值,再结合已知的有雾图像反演计算求出无雾图像,利用MATLAB仿真加以实现,并将改进去雾算法与直方图均衡去雾算法、Retinex去雾算法进行对比。在AX301芯片上设计搭建基于FPGA的图像视频去雾处理平台,包括图像采集模块、图像存储模块、视频传输与显示模块,为实现视频图像实时去雾处理提供硬件平台。最后,对改进算法进行基于FPGA平台的设计。使用Quartus II软件结合硬件描述语言Verilog实现平台的各个功能模块,并进行时序分析,利用Quartus II内嵌Model Sim对重要子模块进行仿真验证,得出电路图与仿真时序图,并对比本文去雾算法与直方图均衡去雾算法和Retinex去雾算法硬件实现结果,实验结果表明,本文去雾算法FPGA硬件平台资源消耗较小,客观指标与MATLAB处理结果相符,可有效实现有雾图像视频的实时去雾处理,提高去雾算法的效率,同时实现小型化的去雾设备,从而能将去雾设备类比应用到更多场景中。更多还原

【Abstract】 With the continuous development of computer vision systems,digital image video is widely used in our daily lives.However,under the influence of severe weather such as smog,when capturing images and videos,atmospheric particles in smog can absorb and scatter atmospheric light,causing serious degradation of the captured image and video,resulting in phenomena such as blurred image information and reduced contrast,which can have an impact on daily use such as traffic monitoring.Therefore,it is necessary to perform fog removal processing on such image videos to restore clear images.This article will improve the dark channel prior algorithm,combine hardware FPGA(Field Programmable Gate Array)to remove fog from image video,and utilize the parallel computing characteristics of FPGA to accelerate image video defogging.Firstly,this thesis investigates and analyzes the relevant research fields and research status of domestic and foreign defogging processing algorithms and their implementation on FPGA,analyzes various research algorithms and platforms in the entire image defogging field,and obtains the reasons for degradation of image video in foggy weather.Secondly,a physical model is established,and an improved windowed weighted prior adaptive threshold defogging algorithm for dark channel is proposed.Weighted fusion of initial dark channel values and minimum filtered dark channel values,setting the threshold value to obtain atmospheric light,introducing a correction factor to correct the value of transmittance,and then combining the known fog image inversion calculation to obtain a fog free image,which is realized by MATLAB simulation.The improved defogging algorithm and histogram equalization defogging algorithm are combined Retinex defogging algorithm for comparison.Design and build an FPGA based image and video defogging processing platform on the AX301 chip,including image acquisition module,image storage module,video transmission and display module,to provide a hardware platform for real-time video image defogging processing.Finally,the improved algorithm is designed based on the FPGA platform.Use Quartus II software combined with hardware description language Verilog to implement each functional module of the platform,and carry out timing analysis.Use Quartus II embedded Model Sim to simulate and verify important sub modules,and obtain the circuit diagram and simulation timing diagram.And compare the hardware implementation results of the defogging algorithm in this thesis with the histogram equalization defogging algorithm and Retinex defogging algorithm.The defogging algorithm in this thesis consumes less resources on the FPGA hardware platform,The objective indicators are consistent with the results of MATLAB processing,which can effectively achieve real-time defogging of foggy image videos,improve the efficiency of defogging algorithms,and achieve miniaturized defogging devices,thus enabling the analogy of defogging devices to be applied to more scenes.更多还原