| Video and images are the main carrier for information interaction in modern society.Humans rely on eyes to obtain visual information and influence their judgment of the brain,while intelligent devices rely on visual sensors to obtain environmental information.The atmospheric particles in haze environment have a serious scattering effect on the optical path,making it difficult to obtain a clear image,which introduces hidden dangers.With the development of new technologies such as 5G,the Internet of Things and Autonomous Driving,more intelligent devices will work with vision system,and the acquisition of high-quality image data will enhance the reliability and stability of the system.Therefore,image dehazing technology has developed rapidly.In this paper,an FPGA-based adaptive image dehazing algorithm has been proposed,taking full account of the sense details,achieving real-time processing capability and providing an effective solution for hardware implementation on edge computing in the future.The main work and achievements of the thesis are summarized as follows:1)An adaptive mechanism for image dehazing algorithm has been established.It mainly includes two points: one is to propose a haze determination strategy based on image saturation which aims to automatically determine whether the current image is a foggy image;the other is to divide the image scene into depth mutation region of non-sky areas,continuous plane region of non-sky areas and sky areas,according to which the algorithm determines the corresponding dehazing strategy.Meanwhile,a filter size adaptive adjustment strategy has been proposed creatively to solve halo effects when dehazing the depth mutation region.2)A general video processing hardware platform based on FPGA core has been designed.One Xilinx's high-performance FPGA named XC7K325 T serves as the main processing core.A DDR3 SODIMM is equipped on the board and the maximum data transmission bandwidth reaches 11.2GB/s.The equipped four-channel HDMI1.4 input interface and four-channel HDMI1.4 output interface can simultaneously send and receive video data.3)The real-time processing design of the video dehazing algorithm on the FPGA has been realized.The SystemVerilog HDL was used to describe the RTL architecture,fully considering clock tree optimization,pipeline design,clock domain crossing design,high fanout signals design,logic multiplexing and parallel implementation of complex operations.Constrained by timing rules,all paths meet the setup time and hold time requirements.The LUT and RAM resource utilization of the project are 35.52% and 41.69%.The power consumption of the hardware platform for 1080 P video real-time processing is 3.397 W.The project has been uploaded to the public website.4)Comprehensive experiments and evaluations on function and performance of the video dehazing algorithm have been carried out.Experiments and evaluations were performed from three aspects: the dehazing effect of the sky area,the dehazing effect of the scene details,and the overall dehazing effect of the image,avoiding the influence of too few evaluation standards and subjective differences on the experiments.Global saturation,global contrast,peak signal-to-noise ratio and structural similarity were utilized in the evaluation of the overall dehazing effect of the image.Meanwhile,the mist image group and the dense fog image group were tested and evaluated separately.50 randomly selected outdoor fog images were tested.The results indicate that the color of the image after dehazing is natural and free of supersaturation.The average values of the ratio increasement on contrast and saturation are 0.309 and 0.994.With the application of the filter size adaptive adjustment strategy,the dehazed image has no serious distortion compared with the original image,and the average structural similarity reaches 0.881.In addition,experiments indicate that the implementation of the dehazing algorithm on the FPGA hardware platform meets the real-time dehazing processing of 1080 P video. |
PDF Full Text Request