Platform Software Design Of Video Analysis Based On Embedded GPU

With the economic development,the amount of video data generated by various industries also increases sharply.The traditional embedded video analysis system is designed for the special scenes,which is a lack of expansibility.Therefore,it is of great engineering significance to study a video analysis platform that has the capability of real-time processing and can be applied to a variety of analysis scenarios.This thesis develops platform software of video analysis based on embedded GPU,which is designed to extend the operation of various video analysis algorithms.The thesis achieves real-time receiving,decoding,and display of HD video,and designs the general framework of video analysis algorithm,which can be expanded the specific video analysis algorithm The platform software runs on the Tegra X1.According to the software's functions,it is divided into four modules:receiving module,decoding module,video analysis algorithm framework module and display module.The receiving module requests the video data via the real-time streaming media protocol.The decoding module constructs a video decoding pipeline through the Gstreamer framework,and uses the hardware decoding to improve the decoding efficiency.The algorithm framework module defines a common set of algorithm execution flow and accelerates the common image processing methods by using GPU parallel processing to facilitate the access of special algorithms.The display module utilizes the framebuffer technology,a double display buffer mechanism to write analyzed data to the corresponding buffer so as to realize the video display function.The system test results show that the video analysis platform developed in this paper can meet real-time requirements.The algorithm framework can be extended to various scenarios.Therefore,it has good engineering research significance and application value.