Research On Optimization And Parallelization Of Image Video Edge Detection Algorithm Based On GPU

With the booming era of "Big Data and Internet Plus",industries are increasingly required to quickly calculate massive amounts of data.Especially in the field of high-definition image computing,traditional repeated calculations are required for each pixel in the image.For example,many fields need to be the image processed to perform the next operation,and the speed of calculating also affects the overall computational efficiency,which is more prominent in the video field.The calculation speed of each high-definition image affects the processing speed of the entire video.Therefore,how to design parallel algorithms for traditional algorithms has become a hot issue to be solved,and images are the basic elements of video.(1)Firstly,the problem of unclear edge in the edge image calculated by Prewitt is proposed according to the neighborhood continuity of the function to compensate the clear edge area and reduce the blurring edge range to reduce the blurring of the blurred edge to the clear area and to be practical.The complexity of time and space complexity are analyzed,and the research direction of parallel optimization is given.Secondly,the Prewitt parallel algorithm and optimization scheme are considered and designed.Combining the characteristics of CUDA programming frame and the analysis flow and data processing characteristics of the algorithm,the algorithm is computationally intensive and data intensive,which improves the efficiency of the algorithm in the GPU processing core and improves the CPU-GPU collaborative computing efficiency,making the parallel design the efficiency of the program has been greatly improved.(2)At the theoretical level,according to the background with similar characteristics,the Vibe algorithm used in video is improved to the image edge detection field by constructing a partial neighborhood background model and the initial background model.Specifically,the idea of constructing a partial neighborhood background model to avoid detecting too many unrelated edges is proposed.From the edge detection of moving targets,a method of sampling from partial neighborhoods and the whole domain is proposed.On the one hand,it fixes the common problem that the traditional Vibe algorithm can’t eliminate "ghosting";on the other hand,avoids detecting too many background edges,and realizes the detection of edge points only for the edge of the moving foreground.Other fields used this algorithm and the prepossessing stage of the algorithm is used.Secondly,in order to better highlight the practicability of the improved algorithm,combined with the CUDA programming frame under the GPU platform,a parallel algorithm is designed for some links.