Parallel Accelerated Implementation Of Image Dehazing Algorithm Based On OpenCL

Computer vision system has been widely used in all aspects of life,such as traffic monitoring,driving records,satellite maps,security monitoring,etc.,and provides effective solutions for many life problems.However,the image acquisition in the computer vision system is susceptible to weather,and the images obtained in foggy weather are not clear,which seriously affect the acquisition of effective information.Therefore,the image dehazing research has great application value.The effective image dehazing algorithms have large computational cost and long calculation time problems,which seriously hampers the practical application of image dehazing algorithms.This thesis takes the haze removal using dark channel prior algorithm as the research object and improves the effect of the image dehazing by optimizing and improving the algorithm,then using OpenCL technology implements the algorithm in parallel on the CPU+FPGA heterogeneous system to improve the implementation speed of the algorithm.Finally,the efficiency of FPGA-based OpenCL application is studied,and further enhance the algorithm to achieve speed,through a variety of accelerate optimization methods.The main work of the paper is as follows:(1)As the value of the atmospheric light intensity calculated by the haze removal using dark channel prior algorithm is larger,the image after the haze removal is darker.This thesis uses the mean value of all the points which size arranged in the top 0.1% points of the dark channel image as the atmospheric light intensity value,improves the algorithm,and makes the fog-free image more bright and natural,the haze removal effect is more better.(2)Aiming at the time consuming problem of calculating the atmospheric light intensity value in the haze removal using dark channel prior algorithm,this thesis adopts a fast implementation method based on statistics,uses the array statistics and records the number of gray values in the dark channel image,then calculate the sum of the first 0.1% of the points in the array by gray value from large to small,and then calculate the average to get the atmospheric light intensity value,greatly improving the speed of the calculation.(3)Aiming at the time consuming problem of the algorithm,this thesis using OpenCL technology implements the algorithm in parallel in the CPU+FPGA heterogeneous system,and improves the implementation speed of the algorithm.(4)Because different FPGA configuration states have different effects on the execution efficiency of OpenCL application,this thesis studies the acceleration optimization method of OpenCL application on FPGA,and improves the implementation speed of the algorithm greatly by setting the appropriate working group size,memory access optimization,loop expansion and improving the throughput,several FPGA-side optimization methods.