【作者】 蒋宇；

【导师】 张春田；

【作者基本信息】 天津大学 ， 信号与信息处理， 2010， 硕士

【摘要】 随着社会的发展,全世界对公共安全需求日益提高,智能视频监控系统正越来越多地应用于国民生活的许多方面,感知环境中动态视觉信息并进行相关处理己成为计算机视觉的一个重要研究方向。本课题主要研究并实现一个基于DSP的运动目标检测系统,它是以摄像机在固定的情况下拍摄的视频图像序列作为研究对象,并最终达到实时性要求的视频监控系统。在算法方面,文中首先讨论了目前应用较多的几种运动目标检测算法,分析了这些常用算法的适用场合和优缺点,然后在此基础之上,根据本文所处理的视频图像的特点以及系统实时性的要求,确定了一套以背景差分算法为核心的运动目标检测方法,其中使用算术平均法和Surendra算法建立初始背景及背景更新,用OTSU法计算分割阈值,最后对背景差分后二值化图像做形态学处理和运动目标标记。在DSP实现方面,本文选用了TI公司的TMS320DM642作为主处理芯片,软件架构上采用了TI的DSP/BIOS多任务操作系统和RF5参考框架,使用C语言完成各个软件模块的程序设计,并按照实时性要求对整个系统进行了优化,最后给出了实验结果和数据。从DSP仿真结果和数据来看,本文的运动目标检测系统在一定条件下能够准确、及时地检测出运动目标,达到了预期的效果。更多还原

【Abstract】 With the development of the society, the demand for public security is getting higher and higher. The intelligent video surveillance system gets the extensive application in many aspects. Perceiving dynamic visual information and processing these information has already become one of the important research areas in computer vision.The main task of this paper is to study and realize a system for motion object detection based on DSP. It is a video surveillance system which meets real-time demand. The objects studied in the system are the videos collected by a stable vidicon. The paper firstly discusses and analyzes the advantage and disadvantage of some common methods in the field of motion object detection, then chooses the background subtraction method as the core of motion object detection according to the traits of video images and the requirements of real-time for the system. We set up initial background and achieve the adaptive update for background by using the arithmetic average algorithm and Surendra algorithm, calculate the threshold value for segmentation by using OTSU algorithm, and at last use the morphologic method to mark motion objects in the binary picture achieved from background subtraction. For DSP realization, the paper chooses TMS320DM642 produced by TI as the main processor. The soft framework is based on TI’s DSP/BIOS multi-task operation system and FR5. The programs of all modules are completed with C language, and the system has been optimized to meet real-time demand. The experimental results show that the moving objects can be exactly and timely detected by the system under the specified occasion.更多还原