Application Research On Color Recognition Of OpenCL And FPGA Heterogeneous Models

With rapid development of scientific technology,the image processing embedded system is widely applied,such as video monitoring,machine vision,video compression and medical image analysis,et.al.,where high computational performances are all required.Owning to strong parallel processing capability of FPGA hardware,the FPGA based heterogeneous computing model provides a sound solution for image processing application.In addition,OpenCL,as a cross-platform software language,provides a novel agile developing tool for FPGA heterogeneous computing.It is largely proved that the OpenCL based FPGA heterogeneous system for high performance image processing has become an active research area.This dissertation presents a novel embedded vision display system based on the technology mentioned above,which constitutes image collection sub-system,OpenCL acceleration sub-system and image display sub-system.The image collection sub-system takes YUV formatted image from USB video camera and stores in its buffer.OpenCL acceleration sub-system is responsible for image data acceleration.While image display sub-system sends the accelerated image to DA chipset with VGA frame format and eventually display it on VGA.It is highlighted that,apart from the hardware development,an improved grey threshold based color recognition algorithm was proposed in this dissertation.In this algorithm,the variable parametric grey value transformation equation model was derived,which outperforms the constant parametric grey value counterpart against its color recognition weakness.Further,by using the OpenCL sub-system,its practical functionality was verified in terms of performance and computational efficiency.It is noted that both the hardware and software developments are implemented on DE1-SOC platform from Altera Ltd.The experimental results demonstrated the full capacity of improved color recognition algorithm for cube color recognition on OpenCL based FPGA heterogeneous system in terms of multi-target color detection,recognition and classification.By employing large number of images with different resolutions and multiple image processing algoritthms with various computational complexity,it is proved that the processing efficiency of OpenCL sub-system outperforms the ARM based software image processing system by large extent,which is able to meet the real-time requirements from engineering perspective.