【作者】 张鹤；

【导师】 吴谨；

【作者基本信息】 武汉科技大学 ， 电路与系统， 2011， 硕士

【摘要】 在现实生活中,运动图像包含有大量有意义的视觉信息。运动目标检测技术作为计算机视觉、运动图像分析、人机交互和智能视频监控技术的基础已广泛应用于军事、计算机辅助设计、航空航天、智能机器人等领域,其检测结果的好坏直接影响到目标定位、跟踪以及行为理解等后续处理。因此运动目标检测技术的研究有很重要的理论价值和实际应用意义。本文主要研究背景相对静止下的智能视频监控系统中运动目标检测算法,主要工作包括:1.介绍了一些图像预处理方面的相关技术,包括:颜色空间模型、图像去噪、图像均衡化、图像边缘检测、图像形态学处理,并对各项技术进行了实验验证。2.简要介绍了图像阈值分割的相关方法。详细介绍了基于阈值分割的帧间差分法和背景差分法两种运动目标检测算法,包括各算法的原理和优缺点,并对各算法进行实验验证。针对两种算法的不足分别提出了改进算法:基于帧间差分和图像边界信息的运动目标检测算法,以及基于IIR滤波器的背景更新运动目标检测算法。另外,简要介绍了另一种常用的运动目标检测算法——光流法。3.为了解决当目标运动较慢或尺寸较小时易出现漏检的问题,在现有运动目标检测方法的基础上,提出了一种将时域和空域结合的运动目标检测方法,即运动信息和标记多尺度分水岭相结合的运动目标检测算法。该方法以时域信息为基础,结合空域的分水岭分割方法可以得到良好的运动目标检测结果,算法具有快速性和鲁棒性。更多还原

【Abstract】 In real life, moving images contain a large number of meaningful visual information. Moving object detection is the base of computer vision, motion, image analysis, human-computer interaction and intelligent video surveillance technology, which is widely used in the military, computer-aided design, aerospace, intelligent robotics and other fields, the test results have a direct impact to the follow-up treatment such as target location, tracking and behavior understanding. So moving object detection technology has a very important theoretical and practical application of significance. This thesis does research on the moving object detection in the background of a relatively stationary video surveillance system, mainly include:1. Describes some aspects of image pre-processing technologies, including: color space model, image denoising, image equalization, image edge detection, image morphological processing, and experimental validation of the technology.2. Briefly describes the image segmentation method, with fully descriptions for two kinds of moving object detection algorithm: frame difference and background difference, which are base of the threshold segmentation, including the principle of two object detection algorithm, besides the advantages and disadvantages between this two, and making the experimental verification of the algorithm. Considering the disadvantages of two object detection algorithms proposes some improved algorithms, which include motion detection algorithm based on Inter-frame difference method and edge information, and background update algorithm based on IIR filter. In addition, briefly introduces the knowledge of optical flow method, which is another commonly used algorithm for moving object detection.3. In order to solving the problems that easily lack to check when the object running slower or the size is smaller, proposes a method of motion detection combining time domain and spatial, base on existing method of moving object detection, which is the motion detection algorithm that combines of motion information and marker multi-measure watershed. This method based on information in the time domain, can get good results of moving object detection that combined with the watershed segmentation method. The result shows that the detection algorithm can detect object accurately and quickly.更多还原