• Optics and Precision Engineering
  • Vol. 32, Issue 5, 752 (2024)
Kun GONG1, Xin XU1, Xiaoqing CHEN2, Yuelei XU1,*, and Zhaoxiang ZHANG1
Author Affiliations
  • 1Unmanned System Research Institute, Northwestern Polytechnical University, Xi'an70072, China
  • 2National Innovation Institute of Defense Technology, Academy of Military Science, Beijing100850, China
  • show less
    DOI: 10.37188/OPE.20243205.0752 Cite this Article
    Kun GONG, Xin XU, Xiaoqing CHEN, Yuelei XU, Zhaoxiang ZHANG. Binocular vision SLAM with fused point and line features in weak texture environment[J]. Optics and Precision Engineering, 2024, 32(5): 752 Copy Citation Text show less
    Framework of binocular vision odometer system integrating point and line features
    Fig. 1. Framework of binocular vision odometer system integrating point and line features
    Line feature triangulation method
    Fig. 2. Line feature triangulation method
    Schematic diagram of binocular line feature triangulation method based on geometric constraints
    Fig. 3. Schematic diagram of binocular line feature triangulation method based on geometric constraints
    Schematic diagram of weak texture dataset collection devices
    Fig. 4. Schematic diagram of weak texture dataset collection devices
    Comparison of line feature extraction algorithms
    Fig. 5. Comparison of line feature extraction algorithms
    Comparison of binocular matching algorithm under weak texture data
    Fig. 6. Comparison of binocular matching algorithm under weak texture data
    Comparison of image construction and positioning effects between improved binocular point line fusion visual SLAM and ORBSLAM2 algorithms
    Fig. 7. Comparison of image construction and positioning effects between improved binocular point line fusion visual SLAM and ORBSLAM2 algorithms
    Trajectories comparison of four algorithms
    Fig. 8. Trajectories comparison of four algorithms

    算法1:基于改进LSD的线特征提取算法

    输入:图像I,阈值T

    输出:线特征集合

    1. 对输入图像应用高斯平滑。

    2. 在图像的每个像素处计算梯度幅度和方向。

    3. 利用区域生长法得到线矩形区域

    4. 遍历图像中的剩余像素:

      a. 使用梯度方向计算局部梯度方向。

      b. 计算最接近局部梯度方向的直线方向。

      c. 将像素分配给与最接近方向匹配的直线。

    5. 输出线矩形区域的NFA<ε的线段。

    6. 遍历图像中的每个直线段:

      a. 计算直线段长度。

      b. 如果直线段长度小于阈值T

       c. 如果直线段满足式(3),则与相邻的直线段合并继续循环;

       d. 如果直线段不与任何其他直线段平行或垂直,则丢弃它。

    7. 返回剩余的直线段。

    Table 1. [in Chinese]
    DatasetLSD+LBDFLD+LBDOurs(无邻域约束)Ours(无区域约束)Ours
    数量精度/%数量精度/%数量精度/%精度/%精度/%
    KITTI/seq 008666595145727881
    KITTI/seq 019171675656708587
    KITTI/seq 029262714962778082
    KITTI/seq 038863735958727982
    KITTI/seq 047969644563737779
    Euroc/VR 018070585151869091
    Euroc/VR 028383516644889193
    EuRoC/MH 019378716057788385
    EuRoC/MH 029971796168728082
    自建弱纹理数据集4572395632879091
    平均值8471635554798385
    Table 1. Comparison between proposed method and line segment matching method based on LBD
    方 法线特征提取线特征匹配总时间
    LSD+LBD34.217.551.7
    FLD+LBD18.515.333.8
    Ours33.17.440.5
    Table 2. Comparison of time consumption of different algorithms
    Kun GONG, Xin XU, Xiaoqing CHEN, Yuelei XU, Zhaoxiang ZHANG. Binocular vision SLAM with fused point and line features in weak texture environment[J]. Optics and Precision Engineering, 2024, 32(5): 752
    Download Citation