基于ZYNQ SoC的嵌入式智能相机设计

发布时间：2018-06-05 23:03

本文选题：嵌入式 + 智能相机　；参考：《哈尔滨工业大学》2017年硕士论文

【摘要】：随着嵌入式技术的蓬勃发展,传统上基于PC平台设计的仪器设备由于其体积大、成本高、硬件升级周期长等缺点,开始逐步被基于嵌入式处理器的解决方案所取代。半导体技术的发展确保了嵌入式处理芯片可以拥有更多的系统资源和优异的处理性能,能满足用户对不同功能接口和处理速度的要求。为此,本文依托于嵌入式技术,设计了一款具有智能化特点和良好交互特性的图像处理设备—嵌入式智能相机。本文选用ZYNQ SoC作为智能相机系统的硬件处理平台,依据系统设计需求,将系统整体功能结构的开发划分到用户空间、内核空间、底层硬件三个层面。在硬件层,对系统硬件功能进行了模块化设计,分别对基于USB和OV7725的双通道图像采集模块、HDMI显示模块进行开发。在内核层,通过移植嵌入式Linux操作系统,进行了底层硬件设备驱动程序的设计,为在用户层操作底层硬件提供了函数接口。在用户层中,对应用于目标追踪和图像分类的两种智能图像处理算法进行移植实现,通过设计基于Qt的图形操作界面,将智能图像处理算法集成进嵌入式智能相机的顶层应用软件系统中,最终将智能相机系统应用于人脸追踪和工业零件缺陷分类检测中。在人脸追踪系统中,人脸识别成功后启动基于时空上下文的实时追踪功能,有效解决了当目标发生遮挡时追踪出错的问题,实时追踪速率8帧/秒。通过构建基于BOW模型的特定工业零件缺陷分类检测系统,为缺陷分类的实现提供了一种新思路,经实际测试,该分类系统对待测物在尺度、角度上的的变化具有较强鲁棒性,分类准确率达80%以上。
[Abstract]:With the rapid development of embedded technology, the traditional instruments and equipments based on PC platform are gradually replaced by the solution based on embedded processor because of its large volume, high cost, long hardware upgrade cycle and so on. The development of semiconductor technology ensures that embedded processing chips can have more system resources and excellent processing performance, and can meet the requirements of users for different functional interfaces and processing speed. Therefore, based on embedded technology, this paper designs an embedded intelligent camera, which has the characteristics of intelligence and good interaction. In this paper, ZYNQ SoC is chosen as the hardware processing platform of the intelligent camera system. According to the system design requirements, the development of the overall functional structure of the system is divided into three layers: user space, kernel space and underlying hardware. In the hardware layer, the hardware function of the system is modularized, and the dual-channel image acquisition module based on USB and OV7725 is developed, and the HDMI display module is developed. In the kernel layer, by transplanting embedded Linux operating system, the design of the underlying hardware device driver is carried out, which provides the function interface for operating the underlying hardware in the user layer. In the user layer, two intelligent image processing algorithms, which are applied to target tracking and image classification, are transplanted, and the graphical operation interface based on QT is designed. The intelligent image processing algorithm is integrated into the top-level application software system of embedded intelligent camera. Finally, the intelligent camera system is applied to face tracking and industrial parts defect classification. In the face tracking system, the real-time tracking function based on temporal and spatial context is activated after face recognition is successful, which effectively solves the problem of tracking errors when the target is occluded. The real-time tracking rate is 8 frames / sec. By constructing a defect classification and detection system for specific industrial parts based on Bow model, a new idea is provided for the realization of defect classification. After practical testing, the classification system is robust to the change of measurement object in scale and angle. The classification accuracy is over 80%.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】