• Acta Optica Sinica
  • Vol. 38, Issue 12, 1215005 (2018)
Jian Liu* and Di Bai**
Author Affiliations
  • Information and Control Engineering Faculty, Shenyang Jianzhu University, Shenyang, Liaoning 110168, China
  • show less
    DOI: 10.3788/AOS201838.1215005 Cite this Article Set citation alerts
    Jian Liu, Di Bai. 3D Point Cloud Registration Algorithm Based on Feature Matching[J]. Acta Optica Sinica, 2018, 38(12): 1215005 Copy Citation Text show less
    Flowchart of point cloud registration
    Fig. 1. Flowchart of point cloud registration
    Characteristic diagram. (a) Empty circle characteristics; (b) maximized minimum angle characteristics
    Fig. 2. Characteristic diagram. (a) Empty circle characteristics; (b) maximized minimum angle characteristics
    Generation process diagram. (a) Insert a new node P; (b) empty circumscribed circle detection; (c) delete the edge AB; (d) form triangles
    Fig. 3. Generation process diagram. (a) Insert a new node P; (b) empty circumscribed circle detection; (c) delete the edge AB; (d) form triangles
    Delaunay triangulation generation chart
    Fig. 4. Delaunay triangulation generation chart
    Registration charts of a seat. (a) Original point cloud; (b) point cloud of seat; (c) traditional algorithm registration; (d) proposed algorithm registration
    Fig. 5. Registration charts of a seat. (a) Original point cloud; (b) point cloud of seat; (c) traditional algorithm registration; (d) proposed algorithm registration
    Contrast diagrams of registration time. (a) Total registration time; (b) initial registration time
    Fig. 6. Contrast diagrams of registration time. (a) Total registration time; (b) initial registration time
    Registration charts of car model. (a) Point cloud data; (b) 3D reconstruction diagram; (c) side initial registration chart; (d) side accurate registration chart; (e) top initial registration chart; (f) top accurate registration chart
    Fig. 7. Registration charts of car model. (a) Point cloud data; (b) 3D reconstruction diagram; (c) side initial registration chart; (d) side accurate registration chart; (e) top initial registration chart; (f) top accurate registration chart
    Threshold influence curves. (a) Relation curves between average error distance and threshold; (b) relation curves between accurate registration time and threshold
    Fig. 8. Threshold influence curves. (a) Relation curves between average error distance and threshold; (b) relation curves between accurate registration time and threshold
    AlgorithmAverage total registration timeAverage initial registration timeAverage accurate registration time
    Traditional algorithm27.05526.9380.117
    Proposed algorithm25.51125.4160.095
    Table 1. Average registration times
    GroupAverage registration error of traditional algorithm /cmAverage registration error of proposed algorithm /cmTotal registration time of traditional algorithm /sTotal registration time of proposed algorithm /s
    10.3370.30127.05525.511
    20.3390.30127.05125.505
    30.3390.30327.04925.501
    40.3430.31127.05825.511
    50.3480.31627.05625.509
    60.3380.30727.05225.507
    70.3350.30127.06325.524
    80.3400.30627.04625.497
    Table 2. Comparison of average error distance and total registration time of multiple registration experiments
    Number of point cloudICP accurate registration parameter
    ThresholdMaximum number of iterationsTransform matrix differenceMean square error
    Source point cloud257170.015001×10-100.1
    Target point cloud50887
    Table 3. Experimental parameters
    AlgorithmRotational translation matrix of initial registrationRotational translation matrix of accurate registration
    Traditional algorithm0.980-0.1540.125-0.2460.1580.987-0.0210.057-0.1200.0400.992-0.1420  0  0  1  0.975-0.1800.127-0.2510.1880.980-0.0600.089-0.1140.0820.990-0.1440  0  0  1  
    Proposed algorithm0.986-0.1660.194-0.0070.1670.970-0.1760.155-0.0100.1770.984-0.0060  0  0  1  0.988-0.1440.054-0.0440.1490.982-0.1190.099-0.0360.1260.991-0.0090  0  0  1  
    Table 4. Point cloud conversion results of different algorithms
    AlgorithmAverage total registration timeAverage initialregistration timeAverage accurateregistration time
    Traditional algorithm46.35846.2100.148
    Proposed algorithm45.03344.8920.141
    Table 5. [in Chinese]
    AlgorithmX-direction rotation angle /radY-direction rotation angle /radZ-direction rotation angle /radAverage error distance /cm
    Traditional algorithm0.0380.5640.2910.564
    Proposed algorithm0.0230.5340.2770.533
    Table 6. Experimental results of different algorithms