Computer Engineering and Applications ›› 2019, Vol. 55 ›› Issue (24): 147-153.DOI: 10.3778/j.issn.1002-8331.1810-0165

Previous Articles     Next Articles

3D Registration Method for MEEM Tracking and Improving ORB Feature Detection

YONG Jiu, WANG Yangping, LEI Xiaomei   

  1. 1.The National Computer Science and Technology Experimental Teaching Demonstration Center, Lanzhou Jiaotong Univeristy, Lanzhou 730070, China
    2.The National Virtual Simulation Experiment Teaching Center of Railway Traffic Information and Control, Lanzhou Jiaotong University, Lanzhou 730070, China
    3.College of Electronic and Information Engineering, Lanzhou Jiaotong University, Lanzhou 730070, China
    4.Gansu Provincial Engineering Research Center for Artificial Intelligence and Graphics & Image Processing, Lanzhou 730070, China
    5.Lanzhou Yuxin Information Technology Co., Ltd., Lanzhou 730070, China
    6.Meteorological Information and Technological Supporting Center, Gansu Meteorological Service, Lanzhou 730020, China
  • Online:2019-12-15 Published:2019-12-11

MEEM跟踪和改进ORB特征检测的三维注册方法

雍玖,王阳萍,雷晓妹   

  1. 1.兰州交通大学 计算机科学与技术国家级实验教学示范中心,兰州 730070
    2.兰州交通大学 轨道交通信息与控制国家级虚拟仿真实验教学中心,兰州 730070
    3.兰州交通大学 电子与信息工程学院,兰州 730070
    4.甘肃省人工智能与图形图像处理工程研究中心,兰州 730070
    5.兰州宇信信息技术有限责任公司,兰州 730070
    6.甘肃省气象信息与技术装备保障中心,兰州 730020

Abstract: Aiming at the drift problem of 3D registration online tracking model in augmented reality system, and the registration failure caused by the time-consuming problem of feature detection algorithm. A 3D registration method based on MEEM tracking and improved ORB feature detection is proposed. Moving object region is tracked by MEEM algorithm. Stable feature points are extracted by using multi-scale space theory when ORB algorithm is used to detect feature points for tracking target position, and the recursive adjustment method of improved decision tree is adopted, and the feature detection parameters are set at the same time, the feature points between adjacent frames are used to match each other. The tracking data set and the cube model generated by OpenGL are simulated. Simulation results show that the improved ORB feature detection algorithm has scale invariance, higher stability and uniform feature distribution for the detection of registered areas, and the error is reduced by about 42% compared with ORB algorithm. The registration method can basically guarantee that the error is less than 7 mm in the running process. The AR system has better real-time performance, accuracy and robustness.

Key words: ORB algorithm, MEEM tracking, augmented reality, tracking registration, registration matrix

摘要: 针对增强现实系统三维注册在线跟踪模型漂移问题,以及特征检测算法耗时问题导致的注册失败。提出一种基于MEEM跟踪和改进ORB特征检测的三维注册方法。通过MEEM算法对移动对象区域跟踪。对跟踪的目标位置采用ORB算法检测特征点时,采用多尺度空间理论提取稳定特征点,并且采用改进决策树的递归调整方式,同时对特征检测参数设置。利用相邻帧之间特征点的匹配关系求得三维注册矩阵;将跟踪数据集与OpenGL生成的立方体模型进行跟踪注册仿真实验。仿真结果表明,改进ORB特征检测算法对待注册区域的检测具有尺度不变性、更高稳定性以及特征分布均匀,误差相比ORB算法降低约42%,该注册方法在运行过程中基本能够保证误差在7 mm以内;使得AR系统具有较好的实时性、精确性和鲁棒性。

关键词: ORB算法, MEEM跟踪, 增强现实, 跟踪注册, 注册矩阵