基于FPGA的低功耗实时Smart Camera平台的小型化实现和双目视觉应用

发布时间：2018-05-14 03:05

本文选题：物联网 + 大数据　；参考：《北京交通大学》2014年硕士论文

【摘要】：随着智能时代的到来,在物联网市场发展和大数据内在需求的推动下,作为视觉传感网中基础节点的智能相机愈发显现出其重要的科研和市场价值。本文在介绍了智能相机基本概念的基础上,针对其独立嵌入式系统的本质和资源受限的特点提出了在智能相机设计研发过程中需要分析和解决的四个问题。然后,通过对这四个问题的逐步分析和解决,以低功耗、高性能、高灵活度和低成本为目标,我们提出并设计了一款基于FPGA的智能相机,并将其命名为Smart-Eyes。目前,Smart-Eyes有两个版本：基础版本Smart-Eyes (V1.0)具备了智能相机的完整框架和基本功能,能进行视频图像的实时采集、存储、处理、传输和显示,分辨率为720x576,帧率为25fps,总功耗大约4.4W。能同时基于DVI接口和以太网接口进行传输和显示。优化版本Smart-Eyes(V2.0)在基础版本之上,通过小型化降低能耗,采用分层结构提高灵活性,实现双目立体视觉系统增强智能化。根据仿真结果显示,对于Middlebury提供的标准测试集Tsukuba,双目匹配的算法速率可以达到60fps,平均误点率约为11.3%,在保证实时性的前提下,以合理的硬件资源实现了具有一定准确率的双目立体匹配,取得了算法执行速度、匹配准确率和系统硬件资源的很好折衷。
[Abstract]:With the arrival of the intelligent age, with the development of the Internet of things market and the internal demand of big data, the intelligent camera, as the basic node in the vision sensor network, has more and more important scientific research and market value. Based on the introduction of the basic concept of intelligent camera, four problems need to be analyzed and solved in the process of intelligent camera design and development are put forward in view of the nature of independent embedded system and the limited resources. Then, by analyzing and solving these four problems step by step, aiming at low power consumption, high performance, high flexibility and low cost, we propose and design a smart camera based on FPGA and name it Smart-Eyes. There are currently two versions of Smart-Eyes: The basic version of Smart-Eyes V1.0) has the complete frame and basic function of the intelligent camera. It can capture, store, process, transmit and display the video image in real time. The resolution is 720x576, the frame rate is 25fps. the total power consumption is about 4.4w. Can be based on both DVI interface and Ethernet interface for transmission and display. Based on the basic version, the optimized version Smart-Eyesi V2.0 can reduce energy consumption by miniaturization, improve flexibility by adopting layered structure, and realize the enhancement of intelligence of binocular stereoscopic vision system. The simulation results show that for Tsukuba standard test set provided by Middlebury, the algorithm speed of binocular matching can reach 60fps. the average delay rate is about 11.3. The binocular stereo matching with certain accuracy is realized with reasonable hardware resources, and a good tradeoff between algorithm execution speed, matching accuracy and system hardware resources is obtained.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TB852.1

【参考文献】