基于FPGA的路径识别图像传感器的设计
|
摘要
基于彩色路径识别的视觉导航方法是当前自动导航小车领域的研究热点和方向。视觉导航是指根据地面路径和被控对象之间的位置偏差控制其运行的方向，因此，地面彩色路径图像的摄取及其识别处理就成为视觉导航系统中的基础和关键。在当前的视觉导航系统设计中，图像处理的硬件平台都是基于通用微处理器，嵌入式微处理器或者DSP进行设计的。这些处理器一个共同的特点就是数据串行处理，而图像处理过程涉及大量的并行处理操作，因此传统的串行处理方式满足不了图像处理的实时性要求。
鉴于微处理器这方面的不足，作者提出一种使用FPGA实现图像识别的并行处理方案，并据此设计一个智能图像传感器。该传感器采用先进的FPGA技术，将图像采集及其显示，路径的识别处理以及通信控制等模块集成在一个芯片上，形成一个片上系统(SOC)。其主要功能是对所采集的彩色路径图像进行识别处理，获得彩色路径的坐标及其方向角，并将处理结果发送给上位机，为自动导航提供控制依据。
本文将彩色路径的识别处理过程划分为三个阶段，第一阶段为颜色聚类识别，以获得二值路径图像，第二阶段为数学形态学运算，用于对第一阶段中获得的二值图像进行去斑处理，第三阶段为路径中心线的定位及其方向角的测量。图像传感器与上位机的通信采用异步串行方式，由于上位机需要控制该传感器执行多种任务，作者定义一种基于异步串行通信的应用层协议，用于上位机对传感器的控制。在图像的显示中，为了弥补图像采集的速率和VGA显示速率的不匹配，作者提出一种基于单端口存储器的图像帧缓冲机制，通过VGA接口将采集的图像实时地显示出来。
根据上述思想，作者完成了系统的硬件电路设计，并对整个系统进行了现场调试。调试结果表明，传感器系统的各个模块都能正常工作，FPGA中的数字逻辑电路能够实时地将路径从图像中准确地识别出来，充分体现了FPGA对路径图像的高速处理优势，达到了设计预期目标，在一定程度上丰富了路径图像识别处理的技术和方法。
Vision navigation based on colorized path recognition is the hotspot and aspect of the research in all the guidable means for AGV. Vision navigation is a way that controls the direction when vehicle moving based on the positional difference between the path and the vehicle, so, incepting and processing the colorized path image is the base and pivotal approach in it. In the current design of vision guide system, general CPU, embedded CPU and DSP are the foundations of the hardware platform. A common characteristic of these processors is that they deal with the data in series, whereas image recognition arithmetic requires a mass of parallel operation, accordingly, the traditional processing in series can not satisfy the demand for real time.
As the shortage of micro CPU in this aspect, the author proposes a scheme of using FPGA to implement the parallel computing, and designs an intelligent image sensor system. The sensor adopts the advanced FPGA chip technology, makes many functional modules into one chip, including the image gathering and displaying, path identification and communication control and so on , which forms a SOC system. The primary function of this sensor is distinguishing the path from the image and computing the coordinate and orientation of the colorized path, and sending out the result to upper computer.
The paper divides the course of image processing into three phases. The first one is obtaining a binary image via color clustering segmentation. The second one is filtering the black spot in the binary image using the tool of mathematical morphology. The third one is measuring the coordinate and direction-angle of the path. The image sensor communicates with the upper computer via UART. Owing to the upper computer needs the image sensor to implement manifold task, the author even defines an application layer protocol, which is applied to control the sensor by upper computer. For matching the speed of image gathering and showing, the author also devises a frame buffer mechanism for image display based on single port memory, and the image will be displayed on a screen through VGA interface in real time.
With the thinking, the author has made a circuit board according as the sensor principle and debugged it. The results show that all the modules perform very well in the working process, and the digital logic circuit in FPGA can identify the path in real time. It incarnates adequately the advantage of image processing based on parallel computing.
鉴于微处理器这方面的不足，作者提出一种使用FPGA实现图像识别的并行处理方案，并据此设计一个智能图像传感器。该传感器采用先进的FPGA技术，将图像采集及其显示，路径的识别处理以及通信控制等模块集成在一个芯片上，形成一个片上系统(SOC)。其主要功能是对所采集的彩色路径图像进行识别处理，获得彩色路径的坐标及其方向角，并将处理结果发送给上位机，为自动导航提供控制依据。
本文将彩色路径的识别处理过程划分为三个阶段，第一阶段为颜色聚类识别，以获得二值路径图像，第二阶段为数学形态学运算，用于对第一阶段中获得的二值图像进行去斑处理，第三阶段为路径中心线的定位及其方向角的测量。图像传感器与上位机的通信采用异步串行方式，由于上位机需要控制该传感器执行多种任务，作者定义一种基于异步串行通信的应用层协议，用于上位机对传感器的控制。在图像的显示中，为了弥补图像采集的速率和VGA显示速率的不匹配，作者提出一种基于单端口存储器的图像帧缓冲机制，通过VGA接口将采集的图像实时地显示出来。
根据上述思想，作者完成了系统的硬件电路设计，并对整个系统进行了现场调试。调试结果表明，传感器系统的各个模块都能正常工作，FPGA中的数字逻辑电路能够实时地将路径从图像中准确地识别出来，充分体现了FPGA对路径图像的高速处理优势，达到了设计预期目标，在一定程度上丰富了路径图像识别处理的技术和方法。
Vision navigation based on colorized path recognition is the hotspot and aspect of the research in all the guidable means for AGV. Vision navigation is a way that controls the direction when vehicle moving based on the positional difference between the path and the vehicle, so, incepting and processing the colorized path image is the base and pivotal approach in it. In the current design of vision guide system, general CPU, embedded CPU and DSP are the foundations of the hardware platform. A common characteristic of these processors is that they deal with the data in series, whereas image recognition arithmetic requires a mass of parallel operation, accordingly, the traditional processing in series can not satisfy the demand for real time.
As the shortage of micro CPU in this aspect, the author proposes a scheme of using FPGA to implement the parallel computing, and designs an intelligent image sensor system. The sensor adopts the advanced FPGA chip technology, makes many functional modules into one chip, including the image gathering and displaying, path identification and communication control and so on , which forms a SOC system. The primary function of this sensor is distinguishing the path from the image and computing the coordinate and orientation of the colorized path, and sending out the result to upper computer.
The paper divides the course of image processing into three phases. The first one is obtaining a binary image via color clustering segmentation. The second one is filtering the black spot in the binary image using the tool of mathematical morphology. The third one is measuring the coordinate and direction-angle of the path. The image sensor communicates with the upper computer via UART. Owing to the upper computer needs the image sensor to implement manifold task, the author even defines an application layer protocol, which is applied to control the sensor by upper computer. For matching the speed of image gathering and showing, the author also devises a frame buffer mechanism for image display based on single port memory, and the image will be displayed on a screen through VGA interface in real time.
With the thinking, the author has made a circuit board according as the sensor principle and debugged it. The results show that all the modules perform very well in the working process, and the digital logic circuit in FPGA can identify the path in real time. It incarnates adequately the advantage of image processing based on parallel computing.
引文
[1]李碧春 李进．基于视觉导航的自动引导车设计．汽车科技第5期，2006
[2]吴鹏．基于DSP的机器人视觉导航系统．自动化博览，2006
[3]朱军 叶庆泰．基于DSP的AGV视觉导向系统研究．计算机测量与控制，2003
[4]思维科技 胡小锋 赵辉编著．Visual C++／MATLAB图像处理与识别实用案例竞选．北京：人民邮电出版社，2004
[5]Uwe Meyer-Baese著．刘凌胡永生译．数字信号处理的FPGA实现．北京：清华大学出版社，2002
[6]OmniVision Serial Camera Contrls Buss(SCCB)．Functional Specification．V2．1．OmniVision，2003
[7]刘晋 李亦蔚编著．微型计算机原理及应用．大连：大连理工大学出版社，1999
[8]康华光主编．电子技术基础模拟部分(第四版)．北京：高等教育出版社，2000
[9]康华光主编．电子技术基础数字部分(第四版)．北京：高等教育出版社，2000
[10]OV9650 Color CMOS SXGA(1．3 MegaPixel)CameraChip Implementation Guide．V1．1．OmniVision，2004
[11]Spartan-3 FPGA Family：Complete Data Sheet．XILINX，2004
[12]Platform Flash In-System Programmable Configuration PROOMS．V2．6．XILINX，2005
[13]CYTC1051DV33 8-Mbit(512KX 16)Static RAM．CYPRESS
[14]FMS3818 Triple Video D／A Converter 3×8bit，180Ms／s．Fair Child Semi Conductor，2001
[2]吴鹏．基于DSP的机器人视觉导航系统．自动化博览，2006
[3]朱军 叶庆泰．基于DSP的AGV视觉导向系统研究．计算机测量与控制，2003
[4]思维科技 胡小锋 赵辉编著．Visual C++／MATLAB图像处理与识别实用案例竞选．北京：人民邮电出版社，2004
[5]Uwe Meyer-Baese著．刘凌胡永生译．数字信号处理的FPGA实现．北京：清华大学出版社，2002
[6]OmniVision Serial Camera Contrls Buss(SCCB)．Functional Specification．V2．1．OmniVision，2003
[7]刘晋 李亦蔚编著．微型计算机原理及应用．大连：大连理工大学出版社，1999
[8]康华光主编．电子技术基础模拟部分(第四版)．北京：高等教育出版社，2000
[9]康华光主编．电子技术基础数字部分(第四版)．北京：高等教育出版社，2000
[10]OV9650 Color CMOS SXGA(1．3 MegaPixel)CameraChip Implementation Guide．V1．1．OmniVision，2004
[11]Spartan-3 FPGA Family：Complete Data Sheet．XILINX，2004
[12]Platform Flash In-System Programmable Configuration PROOMS．V2．6．XILINX，2005
[13]CYTC1051DV33 8-Mbit(512KX 16)Static RAM．CYPRESS
[14]FMS3818 Triple Video D／A Converter 3×8bit，180Ms／s．Fair Child Semi Conductor，2001