• Laser & Optoelectronics Progress
  • Vol. 56, Issue 16, 161002 (2019)
Xiaosong Shi*, Yinglei Cheng, Zhongyang Zhao, and Xianxiang Qin
Author Affiliations
  • Information and Navigation College, Air Force Engineering University, Xi'an, Shaanxi 710077, China
  • show less
    DOI: 10.3788/LOP56.161002 Cite this Article Set citation alerts
    Xiaosong Shi, Yinglei Cheng, Zhongyang Zhao, Xianxiang Qin. Point Cloud Classification Algorithm Based on IPTD and SVM[J]. Laser & Optoelectronics Progress, 2019, 56(16): 161002 Copy Citation Text show less
    Original point cloud data sample
    Fig. 1. Original point cloud data sample
    Flow chart of IPTD filtering algorithm
    Fig. 2. Flow chart of IPTD filtering algorithm
    Diagram of points to TIN
    Fig. 3. Diagram of points to TIN
    Flow chart of ground point classification algorithm
    Fig. 4. Flow chart of ground point classification algorithm
    Classification results of ground points. (a) Bungalow area; (b) building area
    Fig. 5. Classification results of ground points. (a) Bungalow area; (b) building area
    Elevation distributions of point clouds. (a) Original point clouds; (b) normalized point clouds
    Fig. 6. Elevation distributions of point clouds. (a) Original point clouds; (b) normalized point clouds
    Assessment results of different features
    Fig. 7. Assessment results of different features
    Comparison of boundary classification results. (a) Rough classification result of bungalow area; (b) rough classification result of building area;(c) fine classification result of bungalow area; (d) fine classification result of building area
    Fig. 8. Comparison of boundary classification results. (a) Rough classification result of bungalow area; (b) rough classification result of building area;(c) fine classification result of bungalow area; (d) fine classification result of building area
    Classification results of different algorithms. (a) Artificial classification results; (b) classification results of traditional SVM; (c) classification results of NN-SVM; (d) classification results of proposed algorithm
    Fig. 9. Classification results of different algorithms. (a) Artificial classification results; (b) classification results of traditional SVM; (c) classification results of NN-SVM; (d) classification results of proposed algorithm
    CategoryClassification accuracy ofunnormalized non-ground pointsClassification accuracy of normalized non-ground points
    Using all characteristicsUsing selected characteristics
    Vegetation84.789.990.2
    Building86.590.591.6
    Artificiality48.377.377.1
    Table 1. Classification accuracy of different feature combinations%
    CategoryResults of roughclassificationResults offine classification
    Vegetation77.680.3
    Building69.483.1
    Table 2. Classification accuracy of boundary region%
    AlgorithmClassification accuracyof every category /%Overall classificationaccuracy /%Time /s
    Traditional SVMGround79.279.6397
    Vegetation81.2
    Building78.5
    Artificiality45.1
    NN-SVMGround86.587.2216
    Vegetation87.4
    Building88.2
    Artificiality50.5
    Proposed algorithmGround92.792.6364
    Vegetation91.6
    Building93.3
    Artificiality77.1
    Table 3. Classification accuracy of different algorithms
    Xiaosong Shi, Yinglei Cheng, Zhongyang Zhao, Xianxiang Qin. Point Cloud Classification Algorithm Based on IPTD and SVM[J]. Laser & Optoelectronics Progress, 2019, 56(16): 161002
    Download Citation