Software And Hardware Co-Design Of Haze Removal Using Dark Channel Prior Based On Zynq

The haze weather will cause poor contrast and color distortion of image acquired by visual equipment,affecting the acquisition of image feature information of the drone aerial photography,intelligent driving,military remote sensing system and so on based on visual images,and then affects the reliability.The paper analyzes the formation process of foggy images,integrate image acquisition,haze detection and dark channel prior dehaze algorithm processing on Zynq SoC platform.Based on the high-level synthesis technology,divide the function of the dark channel prior algorithm,and the ARM is used to process the relatively complex algorithm,and the FPGA is used to achieve computationally intensive and parallelizable operations to improve the execution efficiency of the algorithm.The main work of this paper is:(1)Design hardware-accelerated IP cores for dark channel prior algorithms in the FPGA section,including:calculation of dark channels,calculation of coarse transmission maps;Guided filtering refines the transmittance,and uses a steering map,a foggy image,and a product of the two products to parallelize the data to accelerate the algorithm;Considering the large amount of steering filtering calculation,the joint bilateral filtering with edge-preserving function is selected for comparison;Aiming at the problem of brightness after processing in the sky area,combined with the characteristics of fog and sky,a sky segmentation method based on brightness and gradient is proposed,.and the brightness segmentation and gradient segmentation data parallel processing acceleration algorithm is used to further optimize the segmented sky region;Recovery of fog-free images,using fixed-point calculation and pipeline processing to optimize hardware processing;By converting to the HSV color space,appropriately increasing the brightness and then switching to the RGB space output,the problem of darkness of the image after defogging is improved.(2)Implement functional modules of the dark channel prior algorithm in ARM,including:combining the characteristics of the dark channel image of the foggy day,and designing the automatic detection module of the fog based on the dark channel;For the distortion of the image caused by the inaccuracy of the atmospheric light value selection,the quadtree decomposition algorithm is used to optimize the atmospheric light value acquisition;Adaptive contrast enhancement is used to adjust the contrast of the image to enhance image visibility for unnatural contrast after defogging.(3)Build a Linux-based embedded processing system and integrate hardware acceleration IP into the SoC of ARM+FPGA structure,design the driver of the data handling module VDMA and the hardware IP,and design human-computer interaction program between hardware and software,and then build a complete dark channel prior defogging platform.Get quantitative evaluation and analysis of the real-time and effects of the system in different scenarios.The dark channel prior defogging system designed in this paper can realize real-time dehazing processing on video images with resolution of 640*480,and achieve good results.It also has low power consumption and better portability.