• Opto-Electronic Engineering
  • Vol. 50, Issue 2, 220148 (2023)
Jinze Xia, Haoming Sun, Shenghui Hu, and Dongtai Liang*
Author Affiliations
  • School of Mechanical Engineering and Mechanics, Ningbo University, Ningbo, Zhejiang 315000, China
  • show less
    DOI: 10.12086/oee.2023.220148 Cite this Article
    Jinze Xia, Haoming Sun, Shenghui Hu, Dongtai Liang. 3D laser point cloud clustering method based on image information constraints[J]. Opto-Electronic Engineering, 2023, 50(2): 220148 Copy Citation Text show less
    Flow chart of 3D laser point cloud clustering algorithm constrained by image information
    Fig. 1. Flow chart of 3D laser point cloud clustering algorithm constrained by image information
    Preprocessing of point cloud data. (a) Before processing; (b) After processing
    Fig. 2. Preprocessing of point cloud data. (a) Before processing; (b) After processing
    Ground segmentation. (a) Groud points; (b) Non-groud points
    Fig. 3. Ground segmentation. (a) Groud points; (b) Non-groud points
    Sensor coordinate system
    Fig. 4. Sensor coordinate system
    YOLOv5 network structure diagram
    Fig. 5. YOLOv5 network structure diagram
    Schematic diagram of detection frame constraint point cloud
    Fig. 6. Schematic diagram of detection frame constraint point cloud
    Cluster centroid selection graph
    Fig. 7. Cluster centroid selection graph
    Experimental hardware platform and experimental scene
    Fig. 8. Experimental hardware platform and experimental scene
    Align timestamp
    Fig. 9. Align timestamp
    LiDAR and camera calibration. (a) Before calibration; (b) After calibration
    Fig. 10. LiDAR and camera calibration. (a) Before calibration; (b) After calibration
    Clustering results of multiple algorithms. (a) DBSCAN; (b) Euclidean Clustering; (c) K-means++; (d) My-method
    Fig. 11. Clustering results of multiple algorithms. (a) DBSCAN; (b) Euclidean Clustering; (c) K-means++; (d) My-method
    Running time of each module of this method
    Fig. 12. Running time of each module of this method
    fxfycxcyk1k2p1p2
    K657.58660.12296.12246.35
    D0.238809−0.6438020.001786−0.024125
    Table 1. Calibration results of internal parameters
    x/mmy/mmz/mmRoll/radPitch/radYaw/rad
    T59.9452.76−14.46−1.5400.031−1.581
    Table 2. Calibration results of external parameters
    Distribution spacing/cmMy-methodK-meansK-means++Euclidean ClusteringDBSCAN
    ωηωηωηωηωη
    20.440.72
    50.460.70
    100.720.88
    150.620.74
    200.540.70
    Table 3. Affects of distribution spacing on the algorithm
    AlgorithmNumber of correctdivisions/numberClusteringaccuracy/%Average timespent/msAverage number of iterations/number
    DBSCAN25870.113.625
    Euclidean Clustering26271.202.517
    K-means21057.071.95112
    K-means++22260.333.37310
    My-method32086.961.1066
    Table 4. Performance comparison of multiple algorithms
    Jinze Xia, Haoming Sun, Shenghui Hu, Dongtai Liang. 3D laser point cloud clustering method based on image information constraints[J]. Opto-Electronic Engineering, 2023, 50(2): 220148
    Download Citation