【作者】 张起；

【导师】 陈东岳；

【作者基本信息】 东北大学 ， 模式识别与智能系统， 2014， 硕士

【摘要】 近年来,随着科学水平的提升以及计算机视觉和人工智能等领域研究的逐步深入,智能移动机器人技术得到了快速发展。移动式视觉机器人跟踪技术是智能移动机器人的关键技术之一,在相关领域有着广泛的应用前景。移动式视觉机器人的自主跟踪系统的设计,一般包含两部分：1)适用于当前移动式视觉机器人平台的实时快速的目标跟踪算法；2)基于目标跟踪结果的机器人运动控制。前者是机器人跟踪行为的决策依据,后者是跟踪行为的具体执行过程。本文以Koala移动机器人作为实验的移动载体,重点研究并提出了一种基于压缩特征的目标快速实时视觉跟踪算法,使其能够在Koala机器人的平台上实现对移动目标的实时跟踪,主要工作包括以下几个方面：(1)调查和分析了课题背景和目标跟踪的研究成果与技术难点,引入了一种基于压缩特征的目标跟踪算法,并着重分析了该算法的不足及其算法内在原因。(2)针对课题任务,重点研究并设计出了一种基于特征自适应选择的改进压缩跟踪算法。本文跟踪算法从特征空间的生成与优劣评估、自适应特征空间的选择、多维高斯模型的目标特征概率估计的建立、分类器的更新方式和搜索策略的调整等方面进行改进,并在公开的跟踪视频数据库的基础上,进行了对比实验,最后从定性和定量角度对实验结果进行了分析,证明了本文算法在跟踪精度和鲁棒性上具有良好的性能。(3)实现对机器人运动控制。考虑到跟踪的实时性和运算效率问题,本文采用了一种基于视觉的运动控制思路,即根据从摄像头获得的目标与机器人位置的相对空间距离信息来动态调整机器人的运动参数,实现了对目标的快速连续跟踪效果。(4)在以Koala机器人为载体的基础上,搭建视觉跟踪机器人硬件系统,并完成串口通信程序设计等硬件调试工作。最终,用本文设计的算法来实验,达到了移动式视觉机器人对目标自主跟踪的任务效果。本文针对Koala机器人作为移动式视觉机器人平台的情形,提出了一个基于改进的实时压缩跟踪算法,并通过视频测试证明了其有效性；在此基础上搭建了完整的基于视觉信息的移动机器人目标自主跟踪系统,较好地实现了对非特定目标的平稳跟踪效果。更多还原

【Abstract】 In recent years, with the promotion of science and computer technology and the intensive research on the field of computer vision and artificial intelligence, intelligent mobile robot technology has been developed rapidly. The target tracking technology of mobile vision robot is one of the key techniques in the research of intelligent mobile robot and it has a wide application prospect in the related fields.The design of autonomous tracking system based on mobile visual robot generally includes two section:one section is, a real-time target tracking algorithm for the current mobile visual robot platform; another section is, robot motion control algorithm that based on the visual target tracking.The former is the decision-making basis of robot tracking behavior, and the latter is the specific executive procedure for tracking behavior.Based on the Koala mobile robot as our experiment mobile carrier, we study and design a target fast tracking algorithm that based on compressive feature. And it can enable tracking effect for moving target in Koala as the mobile visual robot’s environment. Our thesis mainly completed the following work:(1) The thesis surveys and analyses the research results and technical difficulties in target tracking is also introduced. In this thesis, we introduce a real-time compressive tracking algorithm and analyze the shortcomings of the algorithm and the intrinsic reason of this shortcomings.(2) Aiming at our task, we focus on the design of an improved compressive tracking algorithm based on adaptive feature selection. In this thesis, we improve the research from many aspects,such as the feature space generation and evaluation, the adaptive feature space selection, establishment of target feature probability estimation of multidimensional Gaussian model, the update mode of classifier and adjustment of search strategy. And we conduct contrast experiment based on the public tracking video library, and finally we from the perspective of qualitative and quantitative to analyze the experiment result and prove our the tracking algorithm has good performance in tracking accuracy and robustness.(3) The control of robot motion. Considering the real-time tracking and operational efficiency, this thesis designs a kind of motion control method based on visual. We make adjustment according to the relative spatial location information between the camera and the robot’s position to change the motion parameters dynamically, and finally realize the aim of rapid and continuous tracking effect.(4) We set up the hardware system of robot visual tracking based on the Koala robot as the carrier, and complete the design of hardware debug such as serial communication program. Finally, we use the tracking algorithm proposed in this thesis to realize the task of tracking moving target base on mobile visual robot.Aiming at Koala robot as the mobile visual robot platform, this thesis proposes an improved real-time compressive tracking algorithm, and proves the effectiveness by the video; On this basis, we establish the mobile robot autonomous tracking system based on visual information and achieve the stable tracking effect for unspecified target.更多还原