Real-time Implementation Of Video Haze Removal System Based On Dark Channel Prior For Xilinx Zynq-7000

Due to the influence of the atmospheric scattering, the images obtained in hazy and fog have some degradation phenomena, such as poor visibility and contrast, color migration and fidelity. It seriously affects the human eyes and computer vision for identification and analysis of image information, and directly limits the application of outdoor vision systems. Therefore, it has very important theoretical meaning and urgent practical needs to improve the reliability and robustness of the visual system by dehazing the images of outdoor scenes.This paper focuses on the research and analysis of hazy images degradation causes and mechanism. As while we study the physical process of hazy images degradation based on the atmospheric scattering model, and thus restore the clear image without haze. By the research of algorithm about dehazy, we achieve the dehazy process of hazy images based on the dark channel prior and the guided filter. At the same time, we finish the real-time process of video dehazy in the high definition video processing system based on Xilinx Zynq-7000 platform that play an important role in the theory and practice in the application of dehaze. The main contents of this paper are as follows:1. Based on the model methods of hazy-image dehazing, we establish the model of single image dehazing according to atmospheric scatting theory. Meanwhile, in this dissertation we study deeply the dehazing method by using the dark channel prior and refine the coarse transmission map by using the guided filter. By means of MATLAB, we achieve the introduced algorithm of dehazing and adjust the parameters of guided filter. The results show that it has very excellent dahazing effects to handle the single hazy-image based on the dark channel prior and guided filter.2. In the foundation of keeping dehazing effects, we simplify the obtention of dark channel map and the computation of atmospheric light to reduce the amount of calculation in the hardware implementation.3. We design the high definition video processing system based on Zynq-7000 platform. This system completes the acquisition and output of video, and it also includes the embedded Linux system. We can process of some high-speed video, such as dehazing, enhancement methods and multi-channel video splicing.4. For meeting the needs of real-time video dehazing processing in the engineering applications, the improved algorithm of single hazy-image dehazing based on the dark channel prior and guided filter has been implemented with the FPGA hardware. By using the up to down methodology in designing FPGA modules and the hardware description language Verilog, we have completed the implementation of the models combined with the Xilinx embedded IP. At last the functional correctness of single-image dehazing system was tested and verified.5. Finally, the dahazing algorithm has been transplanted and debugged in the high definition video processing. The final results showed that the system can deal with 25 frame video image effectivly per second for the resolution of 720 × 576 images, so it achieved the real time processing in the video surveillance.