智能相机的异构软件框架设计与实现

发布时间：2018-02-25 16:15

本文关键词： DVS6467稀疏表示智能相机机器视觉异构　出处：《西安电子科技大学》2014年硕士论文　论文类型：学位论文

【摘要】：随着社会发展，监控摄像网络日益增多，传统靠人力排查监控视频不能高效及时地处理突发问题。如何使得摄像网络在安防、智能交通、检测生产等领域高效智能工作，对普通数字相机系统提出了巨大的挑战。依托智能相机的机器视觉算法是一种新型的智能检测与识别手段。当今异构体系架构芯片处理速度不断发展，给图像实时检测与识别带来了福音。因此，在异构体系架构上设计实现智能相机的视频处理软件框架，使开发人员在该框架上方便、快速地移植他们期望的机器视觉类算法。本文所述智能相机异构软件框架工作是基于DVS6467系列嵌入式平台，在该平台上移植Motorlona公司定制的嵌入式Linux操作系统。在德州仪器（简称Ti）公司提供的软件开发套件与基本框架基础上，设计与实现了一套完整的智能相机视频处理软件框架，，并且利用多线程、优化等技术手段充分发挥异构多核处理架构的优势。同时，为了更为突出体现异构处理架构的优势，在Intel-i系列PC上基于QT软件开发同构软件处理框架，与异构处理架构形成性能鲜明对照。最后，将航拍目标识别算法应用在本文设计与实现的智能相机视频处理软件框架上，在智能相机框架上测试该识别算法，其实验结果表明本文设计与实现的软件视频处理框架方案是可行的、有效的，且具有很好的移植性，方便快速开发机器视觉算法。
[Abstract]:With the development of society and the increasing number of video surveillance networks, the traditional manual inspection of surveillance video can not deal with unexpected problems efficiently and timely. How to make the camera network work efficiently and intelligently in the fields of security, intelligent transportation, detection and production, etc. The machine vision algorithm based on intelligent camera is a new method of intelligent detection and recognition. Therefore, a video processing software framework for intelligent cameras is designed based on heterogeneous architecture, which enables developers to transplant their desired machine vision algorithms conveniently and quickly. The heterogeneous software framework of intelligent camera is based on DVS6467 embedded platform. The embedded Linux operating system customized by Motorlona is transplanted on this platform. Based on the software development suite and basic framework provided by Texas Instruments, a complete video processing software framework for intelligent camera is designed and implemented. At the same time, in order to give full play to the advantages of heterogeneous multi-core processing architecture, the isomorphic software processing framework is developed on Intel-i series PC based on QT software. Performance contrasts with heterogeneous processing architecture. Finally, the aerial target recognition algorithm is applied to the video processing software framework of intelligent camera designed and implemented in this paper, and the recognition algorithm is tested on the frame of intelligent camera. The experimental results show that the software video processing framework designed and implemented in this paper is feasible, effective, and has good portability, and it is convenient to develop machine vision algorithm quickly.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP311.52;TB852.1

【参考文献】