Design And Implementation Of Real-time Defogging Hardware Accelerator Based On Image Fusion

With the increase of complexity and suddenness of current weather conditions and the development of electronic technology and Internet of Things,the current various traffic application scenarios require a more effective image dehaze system to cope with the increasing risk of potential traffic accidents.At the same time,foggy images severely restrict the efficiency of artificial intelligence in image recognition,furthermore,it restricts the development of autonomous driving.Therefore,in recent years,the dehaze algorithms have increasingly become the focus of researches.However,the current researches about the defogging algorithm are still focused on exploring the balance between the defogging effect and the complexity of algorithms,and there are not many designs of real-time defogging circuits for real world application scenarios.Therefore,this article aims to combine the contrast stretching and edge enhancement algorithms to explore a hardware-based dehaze algorithm based on image enhancement.The main work of the thesis is:(1)This thesis proposes a fusion defogging algorithm based on contrast stretching and guided filtering.we explore the fusion-based defogging algorithm,using histogram equalization with enhanced contrast,linear stretching,and limited contrast stretch as the intermediate process.And boost filtering can efficiently enhance the edges of fog pictures.Finally,the fusion defogging algorithm with limited contrast stretching and high boost filtering based on guided filtering has better efficiency,and the time can be shortened by more than 90% without loss of defogging quality;(2)In this thesis,a FPGA prototype is also implemented based on the proposed algorithm.According to the fusion defogging algorithm using limited contrast stretching and high boost filtering based on guided filtering,we disassemble and optimize the data stream and divide every operation step into different circuit calculation modules,which include grayscale tranforming module,contrast stretching module,box filter module,variance normalization module,guided filter module,high boost filter module and fusion module.Each module is combined into a defogging accelerator.Experiments show that the defogging accelerator can efficiently and accurately complete the defogging work,and can implement the real time processing for FHD images.