基于OpenCL与FPGA异构模式的Sobel算法研究

Research on Sobel Algorithm Based on OpenCL and FPGA Heterogeneous Models

摘要:

随着计算机科学技术的迅速发展，嵌入式领域实时图像处理应用越来越广泛，然而传统硬件因为自身架构导致并行化程度不高，针对在视频监控、机器视觉、视频压缩、医疗影像分析等领域需要对图像进行高性能计算的问题，提出一种以OpenCL软件模型和FPGA异构模式的高性能图像处理解决方案，实现了图像显示和OpenCL加速功能，以Sobel边缘检测算法为研究对象，进行了算法并行性分析，并在系统中运用OpenCL加速内核算法，与基本的ARM平台和OpenCL共享内存加速机制相比较，展开性能测试，对加速效果进行了研究。实验数据表明，使用该系统处理不同分辨率的图像，OpenCL加速子系统的处理较基于片上ARM硬核的软件处理，实现相同功能上有100倍左右的性能提升。

Abstract:

With the rapid development of computer science and technology, the application of real-time image processing in embedded field is becoming more and more extensive. However, traditional hardware has not been parallelized because of its own structure. Referring to the issue that high performance image processing solution required on video surveillance, machine vision, video compression, medical image analysis, the solution based on OpenCL software model and FPGA heterogeneous mode is proposed to solve the problem of high performance computing, to realize the image display and OpenCL acceleration, the algorithm is used to analyze the parallelism based on the Sobel edge detection as the research object, using the OpenCL acceleration of the kernel algorithm, which is compared with basic ARM platform and OpenCL shared memory acceleration mechanism, launching performance test, then studying the acceleration effect. It is experimentally showed that the use of the system to deal with different resolution images, OpenCL accelerating subsystem processing will be about 100 times better performance improvement than based on the on-chip ARM hardcore software processing, under achieving the same functions.