The Design And Optimization Of HDR Video Generation System Based On Heterogeneous Platforms

With the rapid development of emerging industries such as live broadcasting,virtual reality,and e-sports,people's demands for visual effects are becoming higher and higher.So the development of high dynamic range imaging technology has also attracted the attention of scholars in these directions.However,about the traditional high dynamic range imaging technology,its application conditions are so restrict that it can not apply to streaming video because obtaining a high dynamic range image consumes too much time.The goal of this thesis is to accelerate the Contrast Limited Adaptive Histogram Equalization algorithm on three different heterogeneous platforms: CPU+GPU,CPU+FPGA and ARM+FPGA by using the Open CL parallel development framework,then build three HDR video generation systems based on the three heterogeneous platforms.Firstly,this thesis studies the heterogeneous parallel computing technology based on Open CL,and discusses the method of parallelization of algorithms based on GPU and FPGA.In addition,the current HDR imaging algorithms are deeply studied and they are compared in their implementation methods,processing speed,parallelism and other aspects.Finally,the limit contrast histogram equalization algorithm is determined as the algorithm that will be paralleled in the thesis.Based on the mature CPU+GPU heterogeneous platform,the contrast contrast histogram equalization algorithm is paralleled.The experimental goal of parallelizing the contrast contrast histogram equalization algorithm and generating a high dynamic range video system is verified.The GPU-based kernel program is further optimized by means of memory access optimization and local memory optimization.Finally,the high dynamic range video generation system under the CPU+GPU heterogeneous platform is implemented.when processing 1024*754 resolution images,it can produces smooth high dynamic range video streams.In this thesis,FPGA is also used as the Open CL acceleration device,and the contrast contrast histogram equalization algorithm is applied to the two heterogeneous platforms: CPU+FPGA and ARM+FPGA.To assist the implementation of parallelization program,a simulation environment of FPGA-based kernel program is builded.The acceleration effect of the two kinds of heterogeneous platforms based on FPGA on the contrast contrast histogram equalization algorithm is verified,and the kernel program of the FPGA platform is optimized by two methods: storage access optimization and data processing optimization.Eventually,the high dynamic range video generation system on the CPU+FPGA heterogeneous platform can produce smooth HDR video streams when it processes 1350*900 resolution images.In order to verify the image `quality of high dynamic range images after parallelization,In this thesis,the image quality of serial and parallel algorithms is compared by means of mean square error,peak signal to noise ratio and structural similarity.The experimental results show that using FPGA to accelerate can achieve greater acceleration without reducing the amount of image information,so the limited contrast histogram equalization algorithm can be applied to the real-time generation of high dynamic range video.