基于DSP的实时智能视频跟踪系统设计
摘要
近年来随着硬件技术的不断发展和分析技术的不断拓展，各种图像识别系统的研究和开发得到了广泛地重视。在图像识别领域，特别是在动态图像序列背景（静止背景成像变化）中，运动目标的识别与跟踪因其广泛的应用前景而成为一个重要的研究课题。
本论文将介绍一种动态背景下运动目标的检测与跟踪系统。该系统的核心算法是基于快速傅立叶变换（FFT）的相位相关配准算法。其工作原理为：首先它通过FFT运算计算出相邻两帧图像的功率谱；其次将两个功率谱进行单位归一化得到相位谱并对相位谱进行互相关操作；经过对相位相关谱进行IFFT运算获得相位相关函数；然后检测这一互相关函数的最高峰及其坐标值，该坐标值就是两帧图像背景部分的偏移量；最后对两帧图像进行二次差分操作，得到运动目标位置。
硬件设计以TI公司的高性价比、高速浮点的DSP TMS320VC33芯片作为主处理器。该芯片具有四级流水线结构，可在单周期内执行一次浮点数乘和一次浮点数加操作。硬件系统的其它部分包括视频解码芯片TVP5145；FLASH程序存储器；SRAM数据缓存器；RS485总线发送控制等电路，与DSP电路共同构成低成本实时智能视频跟踪系统硬件平台。
试验结构表明此系统能够高效实时的检测运动目标并实现对目标的跟踪。
The main study field in the paper is the methodology to check and track moving object in dynamic background. The main algorithm named Phase-Related Matching algorithm based on Fast Fourier Transform is introduced and it has 5 steps. Firstly two frames of images are transformed to their power spectrums with FFT. Secondly the two power spectrums are transformed to unitary values and multiply each other to build phase-related spectrum of two images. Thirdly the phase-related spectrum is transformed to phase-related function with IFFT. Fourthly the highest peak value of the function is detected to get the point’s coordinate which is the two images’ offset. Finally the object’s excursion is calculated through quadric difference. A system based on Image Phase-Related Matching algorithm is also introduced in this paper. High-performance floating-point TMS320VC33 DSP chip is main processor of system’s hardware. The chip has excellent Performance-Price rate and floating-point processing capability. Its pipeline structure is made up of 4 function units, and it can execute one arithmetic logic operation and one multiply operation in a single cycle. The whole process checking and tracking moving object in the system is as follows. The analog video stream created by CCD camera is sent to video decoder TVP5145, which can transform video stream to separate YUV signal. Only Y signal is used in the algorithm, and the signal is transferred to DSP chip to further process. With Phase-Related Matching algorithm and other assistant algorithm, the moving object can be detected. After all kinds of algorithms are fulfilled, the control instruction generated DSP is sent to camera running controller through RS485 serial bus and drive camera to track the object. The experiment result indicates the system can effectively check and track moving object. Moreover it is a better anti-jamming system.
本论文将介绍一种动态背景下运动目标的检测与跟踪系统。该系统的核心算法是基于快速傅立叶变换（FFT）的相位相关配准算法。其工作原理为：首先它通过FFT运算计算出相邻两帧图像的功率谱；其次将两个功率谱进行单位归一化得到相位谱并对相位谱进行互相关操作；经过对相位相关谱进行IFFT运算获得相位相关函数；然后检测这一互相关函数的最高峰及其坐标值，该坐标值就是两帧图像背景部分的偏移量；最后对两帧图像进行二次差分操作，得到运动目标位置。
硬件设计以TI公司的高性价比、高速浮点的DSP TMS320VC33芯片作为主处理器。该芯片具有四级流水线结构，可在单周期内执行一次浮点数乘和一次浮点数加操作。硬件系统的其它部分包括视频解码芯片TVP5145；FLASH程序存储器；SRAM数据缓存器；RS485总线发送控制等电路，与DSP电路共同构成低成本实时智能视频跟踪系统硬件平台。
试验结构表明此系统能够高效实时的检测运动目标并实现对目标的跟踪。
The main study field in the paper is the methodology to check and track moving object in dynamic background. The main algorithm named Phase-Related Matching algorithm based on Fast Fourier Transform is introduced and it has 5 steps. Firstly two frames of images are transformed to their power spectrums with FFT. Secondly the two power spectrums are transformed to unitary values and multiply each other to build phase-related spectrum of two images. Thirdly the phase-related spectrum is transformed to phase-related function with IFFT. Fourthly the highest peak value of the function is detected to get the point’s coordinate which is the two images’ offset. Finally the object’s excursion is calculated through quadric difference. A system based on Image Phase-Related Matching algorithm is also introduced in this paper. High-performance floating-point TMS320VC33 DSP chip is main processor of system’s hardware. The chip has excellent Performance-Price rate and floating-point processing capability. Its pipeline structure is made up of 4 function units, and it can execute one arithmetic logic operation and one multiply operation in a single cycle. The whole process checking and tracking moving object in the system is as follows. The analog video stream created by CCD camera is sent to video decoder TVP5145, which can transform video stream to separate YUV signal. Only Y signal is used in the algorithm, and the signal is transferred to DSP chip to further process. With Phase-Related Matching algorithm and other assistant algorithm, the moving object can be detected. After all kinds of algorithms are fulfilled, the control instruction generated DSP is sent to camera running controller through RS485 serial bus and drive camera to track the object. The experiment result indicates the system can effectively check and track moving object. Moreover it is a better anti-jamming system.
引文
Uno T.Ejiri M, Tokunaga T，A Method of Real-Time Recognition of Moving Objects and Its Application，Pattern Recognition, 1976,8:201-208
李智勇，沈振康，杨卫平，海新，动态图象分析，北京：国防工业出版社，1999
孙仲康，沈振康，数字图象处理及其应用，北京：国防工业出版社，1985
罗森菲尔德，数字图象分析，北京：科学出版社，1987
李在铭，数字图像处理压缩与识别技术，北京：电子科技大学出版社，2000
孙即祥，王晓华，钟山，张帆，史慧敏，模式识别中的特征提取与计算机视觉不变量，北京：国防工业出版社，1999
张雄伟，曹铁勇，DSP芯片的原理与开发应用（第二版），北京：电子工业出版社，2002
李方慧，王飞等，TMS320C6000系列DSPs原理与应用，北京：电子工业出版社，2003
肯耐实，数字图象处理 北京：电子工业出版社，1998
徐立中，数字图像的智能信息处理，北京：国防工业出版社，2000
程佩青，数字滤波与快速傅立叶变换，北京：清华大学出版社，1988
C.D.Kuglin, A.F.Blumenthal, J.J.Pearson, Map Matching Techniques for Terminal Guidance using Fourier Phase Information, SPIE, Vol-186, 1979
How to Begin Development Today With the TMS320VC33 DSP，Texas Instrument，1999
TMS320VC33 Data Manual，Texas Instrument，2002
TMS320C3x User’s Guide，Texas Instrument，1997
TMS320C3x Peripheral Control，Texas Instrument，1992
Code Composer Users Guide，Texas Instrument，1999
TVP5145 Data Manual，Texas Instrument，2002
TVP5031/TVP5040/TVP5145EVM Users Guide，Texas Instrument，2001
RS-422 and RS-485 Application Note，B&B Electronics，1997
Initializing the TVP5145 Video Decoder，Texas Instrument，2002
TMS320C3xC4x Assembly Language Tools Users Guide，Texas Instrument，1995
TMS320C3xC4x Optimizing C Compiler Users Guide，Texas Instrument，1997
TMS320C3xC4x C Compiler Users Guide，Texas Instrument，1996
TMS320C3xC4x Code Generation Tools Getting Started Guide，Texas
Instrument，1994
TMS320C3x Simulator，Texas Instrument，1995
http://www.ti.com
http://www.21ic.com
http://www.atmel.com
http://www.cypress.com
http://www.liumeiti.com
http://www.vchelp.com
D.I.Barnea, H.E.Silverman, A Class of Algorithms for Fast Digital Image Registration, TIEEE, Vol C-21, No: 2, 1972.
李智勇，沈振康，杨卫平，海新，动态图象分析，北京：国防工业出版社，1999
孙仲康，沈振康，数字图象处理及其应用，北京：国防工业出版社，1985
罗森菲尔德，数字图象分析，北京：科学出版社，1987
李在铭，数字图像处理压缩与识别技术，北京：电子科技大学出版社，2000
孙即祥，王晓华，钟山，张帆，史慧敏，模式识别中的特征提取与计算机视觉不变量，北京：国防工业出版社，1999
张雄伟，曹铁勇，DSP芯片的原理与开发应用（第二版），北京：电子工业出版社，2002
李方慧，王飞等，TMS320C6000系列DSPs原理与应用，北京：电子工业出版社，2003
肯耐实，数字图象处理 北京：电子工业出版社，1998
徐立中，数字图像的智能信息处理，北京：国防工业出版社，2000
程佩青，数字滤波与快速傅立叶变换，北京：清华大学出版社，1988
C.D.Kuglin, A.F.Blumenthal, J.J.Pearson, Map Matching Techniques for Terminal Guidance using Fourier Phase Information, SPIE, Vol-186, 1979
How to Begin Development Today With the TMS320VC33 DSP，Texas Instrument，1999
TMS320VC33 Data Manual，Texas Instrument，2002
TMS320C3x User’s Guide，Texas Instrument，1997
TMS320C3x Peripheral Control，Texas Instrument，1992
Code Composer Users Guide，Texas Instrument，1999
TVP5145 Data Manual，Texas Instrument，2002
TVP5031/TVP5040/TVP5145EVM Users Guide，Texas Instrument，2001
RS-422 and RS-485 Application Note，B&B Electronics，1997
Initializing the TVP5145 Video Decoder，Texas Instrument，2002
TMS320C3xC4x Assembly Language Tools Users Guide，Texas Instrument，1995
TMS320C3xC4x Optimizing C Compiler Users Guide，Texas Instrument，1997
TMS320C3xC4x C Compiler Users Guide，Texas Instrument，1996
TMS320C3xC4x Code Generation Tools Getting Started Guide，Texas
Instrument，1994
TMS320C3x Simulator，Texas Instrument，1995
http://www.ti.com
http://www.21ic.com
http://www.atmel.com
http://www.cypress.com
http://www.liumeiti.com
http://www.vchelp.com
D.I.Barnea, H.E.Silverman, A Class of Algorithms for Fast Digital Image Registration, TIEEE, Vol C-21, No: 2, 1972.