【作者】 张艳；

【导师】 郭继昌；

【作者基本信息】 天津大学 ， 电子与通信工程， 2012， 硕士

【摘要】 近年来,对运动目标检测的理论和应用研究越来越多,逐渐成为计算机视觉和视频图像处理等众多学科的重要研究课题。运动目标检测就是在所检测的视频帧中将感兴趣的运动目标与背景图像或其他不感兴趣的运动目标相互区分。然而在复杂环境下,运动目标检测的实用性和鲁棒性却很难实现,这也一直是视频图像处理的难点。在复杂环境下,任何环境的改变都会对运动目标检测的准确性产生影响。因此文中提出了基于固定分布数K或自适应分布数K的广义高斯混合模型与背景减除法相结合的算法对运动目标进行检测。该模型可以灵活地感知环境变化,自适应地处理视频背景模型中环境的变化,如光线渐变、背景扰动和噪声等。同时采用基于RGB彩色模型的改进的SNP阴影检测法对阴影进行检测。由于运动阴影也会被检测成运动物体,所以阴影的判别也极为重要。之后用亮度信息来检测是否发生光线突变,当检测到环境中光线发生突变时用背景重新建模机制迅速解决。本文算法复杂度较高,所以在进行视频读取时采取隔帧扫描进行背景更新。实验结果显示,本算法在去除阴影和解决光照变化的问题上具有一定的鲁棒性,在满足实时性的同时可以准确的感知环境变化并检测出运动目标,具有一定的实用价值。然后将算法移植到TI达芬奇DM6446平台上。DM6446是ARM+DSP的双核架构处理器,本文分析了其双核间通信机制的原理,达芬奇开发过程,及其codec engine和DSP sever的工作原理。达芬奇平台工作于Linux嵌入式操作系统环境,算法通过Linux移植到此平台上进行实验。在进行移植前要对算法进行优化,为了保证实时性对视频图像进行相应的处理并采用隔行、隔点、隔帧的扫描进行运动目标检测。更多还原

【Abstract】 In recent years, moving object detection has been a hot spot in theory and application research. It is also an important branch of the subjects which include video image processing, computer vision and so on. Moving object detection is refers to distinguish between the moving object and the background or other which is not interested in the detection in the image sequence or video. In the complex environment, however, it’s always difficult in video image processing area. It has become the serious obstacle of the practicability and reliability in the video image processing system increasingly.In the complicated surroundings, any changes of the environment will influence the accuracy of the target detection. So it put forward an algorithm that combines the Generalized Gaussian Mixture Model (GGMM) (or Adaptive-K Generalized Gaussian Mixture Model (AKGGMM)) and background subtraction to detect moving objects. The model has a flexibility to perceive environment and model the video background adaptively in the presence of environmental changes (such as radial gradient, shadows, background disturbance and noise etc). At the same time, this article uses the SNP shadow detection method, which is based on the RGB color model. The shadow will also be detected for moving object in the complicated surroundings. Therefore, the shadow detection is also important. Next, it uses the brightness information to detect whether sudden illumination change occurs or not. And when it occurs, the model can resolve it quickly. Because the algorithm is very complex, in order to meet the real-time, this model adopt the principle which is updated every three frame modeling. The experiments show that this algorithm can meet the real-time and detect the target motion accurately.It is the realization of the algorithm based on the TI DaVinci platform. DM6446 is a dual-core processor, which include ARM and DSP. This paper analyzes the principle of the dual-core communication mechanism between ARM and DSP, DaVinci development process and the working principle of the codec engine and DSP sever. DaVinci platform works in embedded Linux operating system. Algorithm is transplanted through the Linux to the platform for experiment. Before the transplant, the algorithm must be optimized.更多还原