Accelerated Research On Image Processing Algorithm Based On ARM-FPGA Multi-core Platform

With the development of science technology and the improvement of economic level,image processing plays an increasingly important role in the field of computer.However,with the increasing time and space complexity of image algorithm,more and more resources are needed,which makes the algorithm implemented in software inefficient,resulting in the slow speed of image algorithm implemented in software.After testing,the speed of SIFT algorithm is only about 0.4f/s under Core(TM)i5-4210 H processor.Therefore,it is particularly urgent to speed up the algorithm,especially in the embedded field.With the commercialization of image processing applications,the implementation of the algorithm by software alone can not meet the real-time requirements of enterprises.FPGA is well known for its excellent parallel processing capability,and has many advantages such as high adaptability,low power consumption and low cost.Therefore,heterogeneous multi-core architecture platforms emerge as the times require,such as Zynq-7000 series platforms developed by Xilinx.The platform combines ARM processor and FPGA logic devices with strong parallel computing ability,especially for image algorithms with rich functions and high complexity.The platform also has high flexibility,which can not only show great advantages in flexible development,but also speed up the embedded development process,which establishes a good foundation for image embedded applications.Through Zynq platform,a complete image processing system can be constructed,which can not only bring fast computing speed through the logic resources of the FPGA,but also take into account the diversity of functions by using the ARM processor,which will bring greater performance improvement to the advanced image processing system under the embedded platform.Taking SIFT algorithm and Sobel algorithm as examples,this paper analyses the calculation amount and I/O amount of the algorithm.Combining with the characteristics of Zedboard development board,the calculation tasks of software and hardware are reasonably planned.Through the bottleneck link of localization algorithm,the methods of pipelining,parallelization,sliding window and reconstruction function are used to accelerate the optimization,which improves the running speed of key modules.At the same time,the hardware platform for SDSo C is built through Vivado.By using SDSoC development environment,the high-speed data path between ARM and FPGA and the hardware acceleration of the two algorithms are realized,and a complete image processing system is constructed.Finally,the acceleration of SIFT algorithm is more than 15 times faster than software method,and the Sobel algorithm is more than 11 times faster than software method.This makes the system not only a general platform for accelerating image processing,but also can be widely used in various fields with the help of high-speed data path between PS and PL terminal,and even achieve commercial purposes.