基于DSP的视频去雾并行处理方法

为提高视频去雾的实时性，结合DSP芯片DM6437的硬件特性，提出了一种基于暗通道模型的去雾并行处理方法。采用乒乓式DMA传输方法优化了暗通道处理的并行数据访问和运算效率，综合运用单指令多数据、数组扩展、查表法等方法建立了环境光图生成、像素校正等核心处理的并行计算模型，重点改进了一种均值滤波算法，其平均处理时间约4ms。最终对分辨率 的视频实现了11.4ms/帧的实时去雾处理，并保证了良好的去雾质量。

Parallel Processing Method of Video Defogging Based on DSP

According to DSP characters of the DM6437 chip, this paper presents a parallel processing method based on the dark channel defogging model to advance real-time of video defogging. Data accessing and data operating in the dark channel processing are concurrently running with ping-pong DMA. Some methods including SIMD, array extending and look-up table are adopted in the parallel computational model of ambient light image generating and pixels adjusting. A kind of mean filtering algorithm is improved, and its average processing time is about 4 Ms. They optimize mean filtering and pixels adjusting on parallelism. Finally videos which resolution ratio is can be real-time processed (11.4Ms/F), and defogging performance is good.