• Laser & Optoelectronics Progress
  • Vol. 56, Issue 22, 221002 (2019)
De Zhang, Guozhang Li, Huaiguang Wang*, and Junning Zhang
Author Affiliations
  • Department of Vehicle and Electrical Engineering, Army Engineering University Shijiazhuang Campus, Shijiazhuang, Hebei 050003, China
  • show less
    DOI: 10.3788/LOP56.221002 Cite this Article Set citation alerts
    De Zhang, Guozhang Li, Huaiguang Wang, Junning Zhang. Pose Estimation Algorithm Based on Combined Loss Function[J]. Laser & Optoelectronics Progress, 2019, 56(22): 221002 Copy Citation Text show less
    ComPoseNet model. (a) ComPoseNet workflow; (b) preprocessing and ComPoseNet architecture; (c) image processing
    Fig. 1. ComPoseNet model. (a) ComPoseNet workflow; (b) preprocessing and ComPoseNet architecture; (c) image processing
    Effect of loss function
    Fig. 2. Effect of loss function
    Geometric rule diagrams. (a) Selection rule schematic; (b) distance d; (c) angle θ; (d) angle α
    Fig. 3. Geometric rule diagrams. (a) Selection rule schematic; (b) distance d; (c) angle θ; (d) angle α
    Effects of different loss functions. (a) Original image; (b) traditional method; (c) proposed algorithm; (d) comparison
    Fig. 4. Effects of different loss functions. (a) Original image; (b) traditional method; (c) proposed algorithm; (d) comparison
    Translation errors of different algorithms
    Fig. 5. Translation errors of different algorithms
    Angle errors of different algorithms
    Fig. 6. Angle errors of different algorithms
    Effects of pose estimation of different algorithms
    Fig. 7. Effects of pose estimation of different algorithms
    Effects of different parameters on translation error
    Fig. 8. Effects of different parameters on translation error
    Effects of different parameters on angle error
    Fig. 9. Effects of different parameters on angle error
    Estimation effect diagrams of different parameter constraints. (a) Object; (b) d; (c) d+θ; (d) d+α; (e) d+θ+α; (f) comparison
    Fig. 10. Estimation effect diagrams of different parameter constraints. (a) Object; (b) d; (c) d+θ; (d) d+α; (e) d+θ+α; (f) comparison
    Target detection effects. (a) Telephone; (b) duck; (c) iron; (d) drill
    Fig. 11. Target detection effects. (a) Telephone; (b) duck; (c) iron; (d) drill
    Loss functionNumber of training roundsTranslation error /mAngle error /(°)Pixel errorAccuracy /%
    Traditional2000.04598.3105.98286.391
    Proposed2000.04257.7315.57691.262
    Table 1. Errors and Accuracies of different loss functions
    AlgorithmNumber oftraining roundsTranslationerror /mAngleerror /(°)
    Traditional2000.04598.310
    SSD2000.04377.925
    YOLO2000.04317.889
    Proposed2000.04257.731
    Table 2. Errors of different algorithms
    Loss functionNumber oftraining roundsTranslationerror /mAngleerror /(°)
    Traditional2000.04598.310
    Proposed (d+θ)2000.04428.125
    Proposed (d+α)2000.04397.984
    Proposed(d+θ+α)2000.04257.731
    Table 3. Errors under different parameters
    De Zhang, Guozhang Li, Huaiguang Wang, Junning Zhang. Pose Estimation Algorithm Based on Combined Loss Function[J]. Laser & Optoelectronics Progress, 2019, 56(22): 221002
    Download Citation