Design Of Video Key Frame Extraction System Based On ZYNQ

In recent years,with the application of surveillance video in all walks of life,the data of surveillance video shows a huge growth.It is particularly important to extract the key information from the massive surveillance video and eliminate redundant information.Therefore,it is of great practical significance to develop an embedded key frame extraction system with high accuracy and good real-time performance.Video key frame extraction technology is mainly to extract key video sequences in video stream and eliminate redundant information.Its purpose is to facilitate the analysis and storage of later video.This paper mainly focuses on the key frame extraction algorithm of surveillance video.It mainly introduces Vibe algorithm and integrated feature extraction algorithm of "color feature + texture feature" based on similarity between base frames.With the help of Zynq-7000 extensible platform of Xilinx Company,it realizes image acquisition and transmission,hardware core acceleration and image display in the programmable logic part.At the same time,in the processing system part PS,transplant QT and Open Cv.With the help of Open Cv's related library functions,some complex algorithms are processed to achieve hardware-software codesign.Compared with other key frame sequences extracted from surveillance video,the complexity and computational complexity of key frame extraction algorithm are not studied in this paper.Instead,we strive to use the FPGA and ARMarchitecture of Zynq-7000 to implement embedded video key frame extraction system.In this paper,the key frame extraction system of surveillance video based on ZYNQ-7000 is studied,which mainly includes the following four aspects:1.Analysis and Simulation of traditional video key frame extraction algorithm.The principle and characteristics of video key frame extraction algorithms including inter-frame difference method,background difference method,optical flow method and Vibe foreground detection method are analyzed.PC platform is used to simulate and test different monitoring video scenes.Through comparative analysis of different scenarios,it is concluded that the traditional key frame video extraction algorithm has advantages in extracting frame sequences for specific scenarios.However,in the face of complex scenarios,the advantages of single algorithm extraction are not obvious.2.A video key frame extraction algorithm based on color and texture features is proposed.Expanding and analyzing the characteristics of color space attributes and LBP texture attributes in representing video image features.Combining these two image feature attributes,a method of extracting key frames from video clips by using HSV color histogram and LBP texture operator to synthesize feature vectors.The process of extracting video key frames using integrated feature vectors is described in detail.3.The hardware platform of ZYNQ video key frame extraction system is built.According to the framework structure of Zynq-7000 embedded scalable platform based on FPGA and ARM,the overall scheme is designed,the hardware interface is configured,and the video image is captured by OV5640 camera module.4.The key modules are designed on Zynq-7000 embedded scalable platform,including the image transmission between ARM and FPGA(VDMA),ARM,and Vivado Hls high-level synthesis tool to realize the comprehensive feature algorithm to extract video key frames.Finally,the experimental results are analyzed and summarized.