Fig. 1. Transformations from the disparity map to U-disparity map
Fig. 2. Several planes in the world coordinate system
Fig. 3. Flowchart of the proposed algorithm
Fig. 4. (a) Gray image of the plaster model; (b) pseudo color disparity map; (c) projection result on the U-disparity map; (d) binary image of (c)
Fig. 5. (a)Binary U-disparity map; (b)noise with large disparity range labeled with red oblique lines
Fig. 6. Scene segmentation results of group 1. (a) Gray image; (b) 3D reconstruction result; (c) labeled regions of segmentation results in the disparity map; (d),(e) preliminary segmentation result; (f) final segmentation result
Fig. 7. Scene segmentation results of group 2. (a) Gray image; (b) 3D reconstruction result;(c) labeled regions of segmentation results in the disparity map; (d)-(f) preliminary segmentation results of the background board and the two models; (g),(h) final segmentation results of the two models
Fig. 8. Scene segmentation results of group 3. (a) Gray image; (b) labeled regions of segmentation results in the disparity map; (c) segmentation region of (e) in the corresponding phase image; (d) segmentation region in the corresponding disparity image of (c); (e)-(g) preliminary segmentation result; (h)-(j) final segmentation result
Fig. 9. Disparity sequence obtained by Realsense D415 depth camera by the proposed method. (a) RGB image; (b) disparity map; (c) labeled regions of segmentation results in the rendered disparity map
Evaluation index | Group 1 | | Group 2 | | Group 3 | ① | ② | ① | ② | ① | ② | Precision | 91.46% | 99.92% | | 98.37% | 99.24% | | 99.12% | 99.17% | Recall | 98.65% | 98.28% | 93.66% | 94.55% | 98.18% | 98.88% | F-score | 94.92% | 99.09% | 96.37% | 96.42% | 97.95% | 98.16% | Jaccard | 90.33% | 98.20% | 92.99% | 93.09% | 95.99% | 95.11% | Conformity | 89.29% | 98.16% | 92.46% | 92.58% | 95.82% | 95.32% |
|
Table 1. Comparison of evaluation data of Fig.6-8