Implementation Of OpenVX System Based On Polymorphic Parallel Machine

Computer vision technology has brought convenience to our life and has become an integral part of everyday life. With the increasing demand for real-time and accuracy, the traditional computing methods have been unable to meet the computational demands of computer vision applications. Parallel computers can provide powerful computing ability and can effectively improve the operating speed of vision applications, so the use of parallel technology is of great significance to acceleration of computer vision applications.2013, Khronos Group released OpenVX 1.0 standard specification for computer vision acceleration. OpenVX is a cross-platform API standard and will become a very important research direction of computer vision.At first, this article gives a detailed analysis to OpenVX and conducts an in-depth study on OpenVX Objects, OpenVX Kernel functions and graph-based execution model. The graph-based execution model includes the construction of graph, the verification of graph, and the execution of graph. Furthermore, this paper studies the polymorphic parallel machine PAAG(Polymorphic Array Architecture for Graphics and image processing). PAAG is a efficient parallel array architecture and offers a variety of communication mechanisms. PAAG supports both SIMD and MIMD modes and these make the parallel computing on PAAG more flexible. Then, I design and implement the OpenVX system based on PAAG. The whole system is divided into two major parts, the host and PAAG The host is mainly responsible for the construction of graph, the verification of graph etc. PAAG is responsible for the execution of the kernel function and offers hardware acceleration. In order to map the graph model to PAAG, I proposed two mapping strategies and and analyse them.Finally, the system has been tested on the PAAG software simulation environment and I analyze the result based on the execution time and speedup. The data shows that the OpenVX system on PAAG can achieve the acceleration of computer vision applications.