Optimization Of Defogging Algorithms And FPGA Acceleration Method Based On Heterogeneous Atmosphere Light Prior

The frequent occurrence of fog and haze weather not only lays hidden dangers to people's health,but also greatly reduces the visibility and visual effects of roads and surrounding living environments,which seriously affects people's judgment of surrounding safety.simultaneously,the smog makes many image collections based on optical imaging disturbed,which leads to the degradation of the image and the loss of detailed information.Therefore,the research work of image dehazing is greatly crucial.Among the many defogging algorithms,a part of the defogging algorithms optimizes the estimation method of atmospheric light value and transmittance based on the physical model of atmospheric scattering,which can reversely restore the clear original appearance of the image.However,there are still some problems with this type of defogging algorithm based on the atmospheric scattering model: First of all,the traditional atmospheric scattering physical model is based on the premise that the atmospheric light is uniform in the ideal state,so that atmospheric light is often estimated as a constant global constant,which is difficult to reflect the complex and changeable atmospheric environment in real scenes;Then,defogging algorithms based on physical models mostly rely on high-performance computing platforms such as general-purpose CPUs or GPUs,which is difficult to meet embedded computing scenarios that have special requirements on volume,power consumption,and running speed.In view of the above problems,this paper optimizes the existing physical modelbased defogging algorithm,and designs an efficient image defogging hardware system by making use of the heterogeneous parallelism and acceleration features of FPGA.The main work is as follows:Firstly,this paper confirms the rationality of the Heterogeneous Atmospheric light through the analysis and research of the existing defogging algorithm based on physical model,and designs a novel method for estimating the heterogeneous atmospheric map based on Heterogeneous Atmospheric Scattering Model(HASM).Meanwhile,this paper presents a refined transmittance algorithm,which aims at the shortcomings of the roughness of the existing transmittance estimation method.Then the subjective visual effects and objective evaluation indicators(PSNR,SSIM,and CIEDE2000)are compared to the defogging ability of the algorithm in this paper and the other four defogging algorithms.The results show that,our algorithm in this paper has achieved good results in both of improving the visual effect and objective comment.Secondly,in view of the characteristics of limited FPGA resources and strong parallel performance,this paper optimizes the improved defog algorithm in parallel and accelerates the design of FPGA.Meanwhile,the implementation method of memory consumption in the algorithm process is replaced equivalently to reduce the consumption of FPGA on-chip resources.By using the high-level synthesis tool,the IP logic of several modules is instantiated,such as fast guide filter,atmospheric light estimation,side window mean filter,and ambient light calculation.Based on the topdown system design method,the hardware demisting system is implemented on the real physical platform.Finally,this paper compares the performance of PC implementation and FPGA accelerated demisting scheme.The experimental results show that,compared with the PC scheme,the FPGA implementation has slightly decrease(the decrease is not more than 5%)in objective evaluation indicators,but it achieves the similar results to PC implementation in subjective visual experience,and the processing speed is increased by 26.4 times.Moreover,the FPGA's solution only takes6.847 ms to process a 512 * 512 fog image,which corresponds to the actual needs of image enhancement in embedded computing environment.