Fig. 1. Framework overview and applications.

Fig. 2. Relative spatial position and geometric relationship between the ToF sensor and the surface element dA.

Fig. 3. Schematic diagram of horizontal inclination angle α and vertical inclination angle β. (a) is the front view of α, (b) is the top view of α, (c) is the front view of β, and (d) is the side view of β.

Fig. 4. Theoretical approximation of the infrared reflected intensity with respect to (a) the landscape and portrait positions and (b) horizontal and vertical inclinations of the surface element.

Fig. 5. Algorithm framework for recognizing object pose and surface material.

Fig. 6. Schematic diagram of acquisition scene and equipment arrangement.

Fig. 7. Example of data preprocessing and contour extraction. (a) Depth map of the scene after mean noise reduction, (b) result of binarization after depth-valued filtering, (c) result after corrosion noise reduction, (d) result after contour extraction, and (e) result after the second binarization operation.

Fig. 8. Example of the point cloud information for (a) the scene and (b) the extracted result.

Fig. 9. Panels (a), (b), and (c) are the surfaces of different target objects.

Fig. 10. Flow chart of the judgment method based on the fitting results of the IR intensity values.

Fig. 11. Active infrared intensity distribution of different materials by GA-BP network fitting. (a) Cardboard, (b) leather, and (c) metal.

Fig. 12. Active infrared intensity distribution of different materials by DBO-BP network fitting. (a) Cardboard, (b) leather, and (c) metal.

Fig. 13. Fits to the active IR intensity distribution of different materials as a function of the depth of the surface element (a) at position (320, 240) and (b) at position (212, 80).

Fig. 14. Active infrared intensity fitting results at different inclination angles of (a) GA-BP network and (b) DBO-BP network.

Fig. 15. Optimization of the fitting intensity for (a) materials that follow the general diffuse reflectance law and (b) high reflectance materials.

Fig. 16. Optimized fits of infrared intensity distributions for different materials based on the data in Figs. 11 and 12. (a) Cardboard, (b) leather, and (c) metal.

Fig. 17. Optimized fit of the infrared intensity of cardboard at different inclination angles based on the data in Fig. 14.

Fig. 18. Schematic illustration of D-AI image data acquisition locations for each material.

Fig. 19. Sets of (a) intensity images and (b) contour images obtained by depth image of the cardboard collected at the same location and different inclination angles.

Fig. 20. (a) Total pixel accuracy and (b) maximum and average deviations of each set of test samples without inclination.

Fig. 21. (a) Total pixel accuracy and (b) maximum and average deviations for each set of test samples with inclination.