Research On Key Technology Of Intelligent Video High-speed Data Processing System Based On UltraScale FPGA

In recent years,video surveillance has been widely applied in various fields.With the continuous expansion of the number of monitoring systems,more and more high-definition,networking and intelligentization of video,the amount of video data is increasing in an explosive manner,forming a large amount of video data.For massive monitoring video data,it is becoming more and more important to improve the transmission efficiency,improve the utilization of resources,intelligent analysis,high-speed processing and transmission on the premise of guaranteeing the transmission quality.As a programmable hardware platform,UltraScale FPGA has this large-scale distributed hardware resource.In addition,the platform has the characteristics of short development cycle,good performance and so on.It is very suitable for the processing of intelligent video high-speed data.In this paper,UltraScale FPGA is used as the hardware development platform to study the key technologies of the intelligent video high-speed data processing hardware platform.The main research contents are as follows:I had come to understand the current research status and deeply study the related technologies of intelligent video data processing,analyze the requirements of intelligent video high-speed data processing system,select the appropriate experimental platform,and design a system of intelligent video high-speed data processing system based on UltraScale FPGA platform.Then,I have carried out in-depth research on the key technology of intelligent video processing system,video capture and transmission,aiming at the situation that one network video data takes up one DMA in the traditional video capture and transmission scheme,which needs a lot of resources.By studying the communication mechanism of AXI-Stream bus protocol,DMA,AXI-Ethernet IP and DDR,this paper designs an arbiter based on multi-channel network interface data transmission,which can realize the concurrent transmission of multi-channel network data through DMA.Under the premise of ensuring the transmission quality,it can significantly improve the resource utilization ratio,and can meet the needs of the intelligent video high-speed data processing system,which can guarantee the transmission quality and have a higher resource utilization ratio at the same time.The key technology of intelligent video data processing system is deeply studied by me.Through the research of intelligent video processing technology,convolutional neural network has excellent performance in video image data processing,and it is suitable to be implemented on FPGA.The algorithm analysis of convolutional neural network shows that it can be parallel independent source,from the aspects of parallelism,storage structure and so on,the network operation structure is designed.The copy_buffer IP core is designed to accelerate the access of video images.The matrix multiplication unit,which has the largest proportion in the convolution neural network,is optimized by using block technology and pipelining techniques,and is multiplexed to multiple layers for parallel accelerated computation.The realization of convolutional neural network is accelerated in UltraScale FPGA,and the speed of intelligent video processing is improved.Finally,I set up the whole experimental platform of the system,and analyze the optimized design of video data acquisition and transmission with the optimization design of video image processing and contrast analysis,and analyze the power,speed and resource occupancy rate.The implementation shows that the hardware system of intelligent video high-speed data processing based on UltraScale FPGA platform is complete and has excellent performance.