| With the persistent advance of science and technology,the vision system is widely used in intelligent transportation,autonomous aircraft,industrial security monitoring,and other fields.In haze and dust weather,there are problems such as fuzzy texture details and weak color information in visual system imaging,which is not conducive to the judgment of image feature information.Limited by the computing capacity,the current general-purpose processors cannot meet the fast restoration requirements of high-resolution degraded images.Therefore,it is of great significance to study a dehazing algorithm with good effect and strong versatility and design a real-time dehazing system with low cost,small size,and excellent real-time performance.In this thesis,the advantages and disadvantages of major hardware acceleration platforms for image dehazing are analyzed and compared,and ZYNQ So C with both parallel computing capabilities and flexible control capabilities is selected as the dehazing algorithm acceleration platform.The principles and effects of various dehazing algorithms are analyzed and compared.Based on the consideration of the efficiency of hardware transplantation and the diversity of applications,the Retinex dehazing algorithm based on the center-surround model is selected as the dehazing system deployment algorithm.Secondly,aiming at the shortcomings of the central-surround Retinex model,an improved Retinex dehazing algorithm is proposed: Bilateral filter and Gamma correction were used to improve the illumination estimation process of the original model,which solved the problems that the original algorithm did not have halo elimination ability and has low accuracy of illumination estimation;Combined with the advantages of Contrast-Limited Adaptive Histogram Equalization algorithm,the ability of the algorithm to enhance details in dense fog is improved;The introduction of color enhancement factor and white balance processing realizes the enhancement of the original color of the image and the correction of color temperature.Through comparative experiments,it is verified that the improved algorithm has a good enhancement effect for degraded images of different degrees,the improved method is feasible and has practical use value.Finally,the parallel and streamlined FPGA transplantation of the improved Retinex dehazing algorithm is realized by using a high-level synthesis method,combined with the algorithm principle and the characteristics of FPGA,the area and speed of the hardware circuit of the dehazing algorithm are optimized: The circuit timing is improved by using the look-up table method instead of the logarithmic calculation in the improved Retinex algorithm;The cyclic structure with special dependencies is combined to reduce the processing delay of the algorithm;Fixed-point is used instead of floating-point to save bit-width resources.The RTL simulation results show that under the operating frequency of 100 MHz,the dehazing circuit can process 1080 p resolution foggy images at a speed of 47 frames per second.Based on the idea of software and hardware co-design,the defogging circuit is packaged into an IP core whose parameters are flexible and adjustable at the ARM end,and the real-time defogging system based on ZYNQ So C is designed,the system resource consumption is less than 35%.The dehazing system is tested to realize the expected real-time image defogging enhancement function. |
