Fig. 1. Diagram of system
Fig. 2. Depth image preprocessing. (a) Raw depth image; (b) shadow noise area; (c) ROI segmentation; (d) reliability mask
Fig. 3. Depth histogram
Fig. 4. Weight setting near depth edge. (a) σ0 neighborhood range; (b) ωm function curve
Fig. 5. TSDF variation
ηTS. (a) Color image; (b)-(d)
ηTS<
γl, low pass filter parameter
γl=0.6, 0.7, 0.8; (e)-(h)
ηTS>
γh Fig. 6. Reference frame FIFO queue
Fig. 7. Model reparation experiment. (a) Model before reparation; (b) model after reparation; (c) experimental environment and setting; (d) peculiar object; (e) model detail before reparation; (f) model detail after reparation
Fig. 8. Models corresponding to different pose results. (a) Experimental setting; (b) KinectFusion; (c) SDF-tracker; (d) proposed method
Fig. 9. Comparison of reconstruction results 1. (a) Experimental environment and setting; (b)camera pose; (c) result obtained by algorithm Ⅰ; (d) result obtained by proposed method
Fig. 10. Comparison of reconstruction results 2. (a) Experimental environment and setting; (b)camera pose; (c)(d) result obtained by algorithm Ⅰ; (e)(f) result obtained by proposed method
Method | Average pose errorafter 100 frames | Average pose errorafter 300 frames | Average pose errorafter 600 frames |
---|
KinectFusion | 32.8 | 38.2 | 62.6 | SDF-tracker | 26.1 | 27.5 | 33.8 | Proposed method | 25.7 | 27.0 | 29.7 |
|
Table 1. Error analysis of pose tracking in experiment Ⅰmm
Method | Average pose errorafter 100 frames | Average pose errorafter 300 frames | Average pose errorafter 600 frames |
---|
KinectFusion | 51.0 | 100.2 | 811.2 | SDF-tracker | 28.9 | 33.4 | 38.8 | Proposed method | 30.5 | 33.2 | 33.7 |
|
Table 2. Error analysis of pose tracking in experiment Ⅱmm