暗通道先验优化的FPGA实时去雾系统

Hardware-optimized Dark Channel Prior Algorithm and Real-time Dehazing System

摘要:

透射率的求取决定了暗通道先验去雾算法的执行效率和去雾效果；在传统的透射率求取中,软抠图和引导滤波计算复杂且耗时长,对算法进行实时处理的移植会导致资源消耗大,内存带宽限制及硬件设计困难；图像中包含亮度值较高的天空区域时,不满足暗通道先验条件致使透射率求取出现偏差及天空部分色彩失真；为解决这些问题,通常做法为对滤波方法进行硬件优化；通过改进滤波窗口大小并引入均值滤波优化透射率图像求解过程,基于亮度阈值差采用天空分割策略计算图像中天空所占比例,根据此比例自适应调整透射率边界值；利用FPGA并行处理优势,将其作为核心控制硬件在Xilinx平台对所优化算法进行设计与实现,在图像传感至显示系统中通过了实时去雾效果验证,实验证明优化算法的去雾效果在主客观评价指标上均优于传统暗通道先验,处理一帧1080P高帧率图像仅耗时38.6ms。

Abstract:

The calculation of the transmittance determines the execution efficiency and dehazing effect of the dark channel prior dehazing algorithm. In the traditional transmittance calculation, soft matting and guided filtering are complex and time-consuming. The real-time processing of the algorithm will lead to large resource consumption, memory bandwidth limitation and hardware design difficulties. When the image contains the sky area with high brightness value, the dark channel prior condition is not satisfied, which leads to the deviation of the transmittance and the color distortion of the sky part. In order to solve these problems, the usual practice is to optimize the hardware of the filtering method; by improving the size of the filtering window and introducing the mean filtering, the transmission image solving process is optimized. Based on the brightness threshold difference, the sky segmentation strategy is used to calculate the proportion of the sky in the image, and the transmission boundary value is adaptively adjusted according to this proportion. Using the advantages of FPGA parallel processing, it is used as the core control hardware to design and implement the optimized algorithm on the Xilinx platform. The real-time dehazing effect is verified in the image sensing to display system. Experiments show that the dehazing effect of the optimized algorithm is superior to the traditional dark channel prior in subjective and objective evaluation indicators. It only takes 38.6ms to process a frame of 1080P high frame rate image.