异构多核低功耗嵌入式视频中间件技术研究

发布时间：2018-05-13 23:04

本文选题：视频 + 嵌入式中间件　；参考：《上海海洋大学》2017年硕士论文

【摘要】：I视频作为现代数据的重要来源,其相关技术的重要性日显突出。尽管目前视频技术在农业、工业、军事、安全、交通等领域已经有较为普遍的应用,但物联网视频技术仍有诸多问题亟须解决,如访问接口不统一、前端计算处理能力弱、不能向用户有针对性地提供数据,多数引用环境中,视频流要经由感知层、传输层到处理层再至应用层,通常伴有着一定的延迟,影响物联网视频监控的实时性与处理效率。为方便视频的转发,减少视频、图像数据处理延迟,提高物联网数据传输与处理效率,本文基于RTSP(Real Time Streaming Protocol)协议及“就地”处理(Process in Field)思想提出了一种嵌入式视频中间件技术,通过低功耗、低成本、高性能的嵌入式异构多核计算系统对现场视频进行采集、处理与转发,本文所做的主要工作包括:(1)调查研究物联网视频监控技术现状,分析其中存在的问题。提出基于RTSP转发与Epiphany低功耗异构多核架构的“就地”处理物联网嵌入式视频中间件解决思路。(2)研究图像视频计算中广泛应用的基础算法(矩阵运算与DCT算法等)在异构多核嵌入式平台上的移植与实现。研究分析了多种主要并行编程模型:Epiphany SDK、MPI、Open MP、Open CL在Epiphany并行异构多核架构上的可编程性和性能。研究分析了核间通信对Epiphany架构图像视频基础算法性能的重要影响。(3)基于Epiphany架构实现了一种VLC+DPM(VLC,Video LAN Client;DPM,Deformable Part Model)的视频转发与模式识别的物联网嵌入式中间件,在研究图像视频计算基础性算法的基础上,分析DPM算法并将其移植到Epiphany架构。通过高性能低功耗的嵌入式多核计算架构,实现物联网前端视频现场处理,完成视频的处理与转发。本文所提技术在功耗只有6.587瓦的嵌入式结点实现视频监控,实验结果表明512*512矩阵乘法的性能达到酷睿i7 2600计算机的处理水平,是一种低功耗、低成本的视频实时中间件技术,为减少视频数据处理延迟,提高数据处理效率提供了有益的借鉴和参考。
[Abstract]:I video as an important source of modern data, its related technology is becoming increasingly important. Although video technology has been widely used in agriculture, industry, military, security, transportation and so on, there are still many problems to be solved in Internet of things video technology, such as the disunity of access interface and the weak computing and processing ability of front end. In most citation environments, video streams pass through perception layer, transmission layer to processing layer and then to application layer, usually accompanied by certain delay, which affects the real-time performance and processing efficiency of Internet of things video surveillance. In order to facilitate video forwarding, reduce the delay of video and image data processing, and improve the data transmission and processing efficiency of Internet of things, this paper presents an embedded video middleware technology based on RTSP(Real Time Streaming Protocol) protocol and "in-place" process in field. Through the low power, low cost, high performance embedded heterogeneous multi-core computing system to capture, process and forward the scene video, the main work of this paper includes: 1) investigating and researching the current situation of the Internet of things video surveillance technology. The existing problems are analyzed. Based on RTSP forwarding and Epiphany low power heterogeneous multi-core architecture, this paper puts forward the idea of "in-place" processing embedded video middleware in the Internet of things (IOT). The basic algorithms (matrix operation and DCT algorithm, etc.) which are widely used in image and video computing are studied. Transplantation and implementation on heterogeneous multi-core embedded platform. The programmability and performance of several main parallel programming models: Epiphany SDK / MPI Open Open Open CL on Epiphany parallel heterogeneous multicore architecture are studied and analyzed. This paper studies and analyzes the important influence of inter-core communication on the performance of the basic algorithm of image and video in Epiphany architecture. (3) based on the Epiphany architecture, a kind of embedded Internet of things middleware for video forwarding and pattern recognition is implemented, which is based on the Epiphany architecture. On the basis of studying the basic algorithm of image and video computing, the DPM algorithm is analyzed and transplanted to the Epiphany architecture. Through the embedded multi-core computing architecture with high performance and low power consumption, the front-end video field processing of the Internet of things is realized, and the video processing and forwarding are completed. The technology proposed in this paper realizes video surveillance at an embedded node with a power consumption of only 6.587 watts. The experimental results show that the performance of the 512 / 512 matrix multiplication is up to the processing level of the Core i7 2600 computer. It is a low power and low cost video real-time middleware technology. It provides a useful reference and reference for reducing the delay of video data processing and improving the efficiency of data processing.
【学位授予单位】：上海海洋大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN948.6;TP391.41

【参考文献】