基于视觉的运动目标检测与跟踪算法的研究与实现

发布时间：2018-03-05 12:29

本文选题：运动目标检测　切入点：mean　出处：《南京理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：近年来,随着机器视觉、智能视频监控等领域的快速发展,运动目标的识别与跟踪技术日渐发展与成熟。针对基于视觉的运动目标跟踪技术,论文提出了一种基于ARM嵌入式平台的运动目标跟踪方案,将嵌入式平台和运动目标跟踪技术结合起来。本论文主要工作和成果如下:(1)研究了三种典型的的运动目标检测算:光流法、帧间差分法和基于混合高斯模型的背景减除法。在各种现有算法的基础上,提出了一种基于Surendra背景更新算法的运动目标检测算法,从二值差分方法、自适应阈值分割和背景更新策略三个方面进行了改进。(2)对目标跟踪算法进行了研究。在两种主流目标跟踪算法:mean shift和粒子滤波算法基础上,提出了一种以粒子滤波算法为框架的改进算法。改进的粒子滤波算法融合了颜色和运动两种特征,采用一种改进的重要性重采样方法,并且把mean shift嵌入到粒子滤波算法中。(3)完成了运动目标跟踪算法的嵌入式设计。构建了一个以MINI5728为核心、搭载USB摄像头的嵌入式平台,完成了软件开发环境的搭建并对嵌入式Linux系统进行移植。在此基础上完成了运动目标跟踪系统的软件设计,包括基于VideoLinux2的图像采集模块设计、基于计算机视觉库OpenCV的运动目标跟踪模块设计以及Qt可视化界面的开发。
[Abstract]:In recent years, with the rapid development of machine vision, intelligent video surveillance and other fields, moving target recognition and tracking technology has been developed and matured. In this paper, a moving target tracking scheme based on ARM embedded platform is proposed. The main work and results of this paper are as follows: 1) three typical algorithms of moving target detection are studied: optical flow method, optical flow method, Based on the existing algorithms, a moving target detection algorithm based on Surendra background updating algorithm is proposed. The adaptive threshold segmentation and background updating strategy are improved. (2) the target tracking algorithm is studied. On the basis of two main target tracking algorithms:: mean shift and particle filter, An improved particle filter algorithm is proposed, which combines color and motion features, and uses an improved importance resampling method. The embedded design of moving target tracking algorithm is completed by embedding mean shift into particle filter algorithm. An embedded platform based on MINI5728 and USB camera is constructed. The software development environment is built and the embedded Linux system is transplanted. On this basis, the software design of the moving target tracking system is completed, including the design of the image acquisition module based on VideoLinux2. The design of moving target tracking module based on computer vision library OpenCV and the development of QT visual interface.
【学位授予单位】：南京理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】