基于DSP与FPGA的运动目标跟踪系统设计

发布时间：2018-04-02 09:36

  本文选题：运动目标跟踪　切入点：并行计算　出处：《哈尔滨工程大学》2016年硕士论文

【摘要】：计算机视觉技术作为自动控制系统的信息获取方式是当前自动化技术领域的一个热点,其主要原因是视觉可以承载更多的信息。但是,视觉信息复杂、数据量大、有效特征提取困难,这给计算机视觉技术的研究带来极大困难,成为目前计算机视觉研究的一个难点。目前,在PC机上实现基于视觉的跟踪技术已经很成熟,但是,由PC机或工作站等实现的视觉系统体积比较大,难以满足灵活性要求。而基于嵌入式的视觉系统可以弥补以上不足,因此越来越受欢迎。基于单一处理器实现的嵌入式算法比较多,而多处理器算法的大部分实现是利用协处理器架构。本文以DSP (数字信号处理器)与FPGA (现场可编程门阵列)相互协同为基础,搭建了一个自动跟踪系统,实现摄像头对运动目标的随动跟踪。一方面,鉴于视频处理中实时跟踪涉及大量的并行运算,适合用FPGA来实现图像的实时滤波、平滑等操作,同时在视频流的采集和处理过程中,存储图像到SDRAM时运用FPGA操作存储器速度更快。另一方面,目标跟踪涉及到大量的复杂运算和迭代算法,在DSP上实现更为容易。本文结合二者的优点,设计用不同的处理器分别处理不同的操作,并行运行相互通信。在FPGA上实现颜色空间的转换、混合高斯模型、形态学二值开运算、中值滤波和线性滤波等图像处理算法和ⅡC时序、UART通讯接口、MCBSP、SDRAM控制器、BT656视频解码等接口;在DSP上实现双分辨率的MeanShift算法用于运动目标的跟踪,并与FPGA进行通信。为了实现摄像头对运动目标的实时瞄准,在ATmega16单片机上实现了 PID算法以控制电机的运动。为了加快系统开发速度,本文首先在PC上论证并设计适合在FPGA与DSP上运行的算法结构,然后将算法映射到本文选定的硬件结构上并通过与PC机通信以调试系统。实验结果表明,本系统对运动目标的随动跟踪具有快速和准确的特点。
[Abstract]:Computer vision technology is a hot spot in the field of automation technology as the information acquisition method of automatic control system. The main reason is that vision can carry more information, but the visual information is complex and the amount of data is large. Effective feature extraction is difficult, which brings great difficulty to the research of computer vision technology, and becomes a difficulty in computer vision research. At present, it is very mature to realize the tracking technology based on vision on PC, but, The visual system realized by PC or workstation is large in volume and difficult to meet the requirement of flexibility. However, the vision system based on embedded system can make up for the above deficiency. So it's getting more and more popular. There are more embedded algorithms based on a single processor. Based on the cooperation between DSP (Digital signal processor) and FPGA (Field Programmable Gate Array), an automatic tracking system is built. On the one hand, because the real-time tracking of video processing involves a large number of parallel operations, it is suitable to use FPGA to realize real-time image filtering, smoothing and other operations. At the same time, in the process of video stream acquisition and processing, it is faster to use FPGA to store images to SDRAM. On the other hand, target tracking involves a large number of complex operations and iterative algorithms. It is easier to implement on DSP. In this paper, different processors are designed to deal with different operations with different processors, running parallel communication with each other. Color space conversion, mixed Gao Si model, morphological binary open operation are realized on FPGA. The image processing algorithms such as median filter and linear filter, and the interface of 鈪，

本文编号：1699787