基于改进混合高斯模型和yolov3算法的运动物体检测与识别-毕业论文.doc

基于改进混合高斯模型和yolov3算法的运动物体检测与识别 Moving object detection and recognition based on improved mixed Gaussian model and yolov3 algorithm 摘 要 近年来，无论是在日常生活的安全防范方面，还是在城市交通、军事上，视频监控技术都得到了广泛运用。视频监控设备一般固定视角采集图像信息，视频监控的连续的图像序列中蕴含着丰富的信息。如何从固定背景的视频序列中提取有用信息则是计算机视觉技术研究领域的热门，其中运动物体检测和物体识别是其中两个研究重点，其中运动物体检测的好坏直接影响后续的识别工作。本文分析了运动物体检测和物体识别两个领域的研究现状，针对固定背景的图像序列，考虑到背景包含摆动的树叶、水波等，在运动物体检测上采用混合高斯建模的背景差分法，该算法能够自适应选择高斯分布的个数，通过对图像的每个像素点进行建模，并根据指定阈值划分前景像素与背景像素，从而得到运动的前景物体，针对学习率固定的问题，通过引入局部学习率改进了现有算法。同时考虑实际运用的实时性要求在物体识别上采用了yolov3算法，该算法能够迅速地检测出待测物体的类别。实验结果表明，混合高斯分布与yolov3算法的结合能够实时提取并识别运动物体。 关键词：运动检测，混合高斯模型，自适应学习率、物体识别，yolov3算法 ABSTRACT In recent years， video surveillance technology has been widely used in both daily life safety， urban traffic， military and so on. Video surveillance equipment generally captures image information at a fixed angle of view， and the continuous image sequence of video surveillance contains rich information. How to extract useful information from fixed background video sequences is a hot topic in computer vision technology research. Among them， moving object detection and object recognition are two of the research focuses， and the detection of moving objects directly affects the subsequent identification work. This paper analyzes the research status of moving object detection and object recognition. For the fixed background image sequence， considering the background containing the oscillating leaves， water waves， etc.， the background difference method using mixed Gaussian modeling is applied to the detection of moving objects. The algorithm can adaptively select the number of Gaussian distributions， and model each pixel point of the image， and divide the foreground pixel and the background pixel according to the specified threshold to obtain a moving foreground object， which is introduced for the problem of fixed learning rate. Local learning rates improve existing algorithms. At the same time， considering the real-time requirements of the actual application， the yol