### Table 3: Stereo 3D reconstruction data from two cameras. First Camera Second Camera Actual

"... In PAGE 40: ... 4.4 Experimental Results Table3 in Appendix B shows the results of twelve 3D points reconstructed from their stereo 2D coordinates based on the decomposition of the perspective transformation matrix M. The Euclidean distance of the recovered coordinates from the actual coordinates is taken as the error.... ..."

### Table 7-1: The computation time for fast interactive 3D model of the single pipeline.

2003

"... In PAGE 64: ...Table7... In PAGE 65: ... Computing time for fast interactive 3D rendering 22% 78% Filtering part Rendering part Figure 7-1: The computation time for fast interactive 3D rendering for data set 256*256. From Table7 -1, we can see rendering part takes longer time than filtering part for large dataset. If we want to reduce the total execution time, we have to not only parallelize the filtering part, but also find a way to reduce the calculation time of the rendering part significantly.... In PAGE 66: ...56*256 21.48 11.42 5.98 2.81 1.39 0.67 Table7 -2: The computation time of the filtering part of the distributed pipeline. Filtering time 0.... In PAGE 67: ...56*256 1.50 1.19 1.08 1.05 1.05 1.06 Table7 -3: The communication time between two parts for the distributed pipeline. Network communication time 0.... In PAGE 67: ...56*256 45.08 23.62 12.37 6.99 4.34 3.14 Table7 -4: The computation time of the rendering part for the distributed pipeline ... In PAGE 68: ...56*256 68.06 36.23 19.43 10.85 6.78 4.87 Table7 -5: The total time of the distributed pipeline ... In PAGE 70: ...56*256 57.30 38.26 19.54 11.68 7.63 5.94 Table7 -6: The total time of the simulator Total time 0.1 1 10 100 N =1 Np = 2 Np = 4 Np = 8 Np = 16 Np = 32 Number of processors and mappers S e c onds 64*64 128*128 256*256 Figure 7-6: The total time of the simulator.... In PAGE 71: ... Using the data from Table7 -6, the speedup is calculated in the following table: N =1 Np = 2 Np = 4 Np = 8 Np = 16 Np = 32 64*64 1 1.... In PAGE 71: ...56*256 1 1.50 2.93 4.91 7.51 9.65 Table7 -7: The speedup of the simulator. Speedup 0.... In PAGE 72: ... Efficiency is bounded by [0, 1]. Using the data from Table7 -7, the efficiencies are calculated in the following table: N =1 Np = 2 Np = 4 Np = 8 Np = 16 Np = 32 64*64 1 0.... In PAGE 72: ...56*256 1 0.75 0.73 0.61 0.47 0.30 Table7 -8: The efficiency of the simulator Efficiency 0.01 0.... In PAGE 92: ...able 4-1: The extraction of the high-added-value features.............................................51 Table7 -1: The computation time for fast interactive 3D model of the single pipeline.... In PAGE 92: ...able 7-1: The computation time for fast interactive 3D model of the single pipeline....64 Table7 -2: The computation time of the filtering part of the distributed pipeline.... In PAGE 92: ...able 7-2: The computation time of the filtering part of the distributed pipeline. ...........66 Table7 -3: The communication time between two parts for the distributed pipeline.... In PAGE 92: ...able 7-3: The communication time between two parts for the distributed pipeline. ......67 Table7 -4: The computation time of the rendering part for the distributed pipeline .... In PAGE 92: ...able 7-4: The computation time of the rendering part for the distributed pipeline ........67 Table7 -5: The total time of the distributed pipeline .... In PAGE 92: ...able 7-5: The total time of the distributed pipeline ........................................................68 Table7 -6: The total time of the simulator .... In PAGE 92: ...able 7-6: The total time of the simulator ........................................................................70 Table7 -7: The speedup of the simulator.... In PAGE 92: ...able 7-7: The speedup of the simulator. .........................................................................71 Table7 -8: The efficiency of the simulator .... ..."

### Table 3: Errors of reconstruction under various camera models and camera positions and

"... In PAGE 15: ...even calibration points to Eq. 36. Six of these points were located on the surface of the sphere and the remaining point was the sphere apos;s centroid. Test results are given in the Table3 . Column 2 gives the mean error averaged over the 20 recovered world points, where error is computed as the Euclidean distance between the actual and the recovered coordinates.... In PAGE 26: ... volume of 162mm 126mm 60mm. The focal length used in the tests was 12.5mm. The reconstruction errors of their method are extracted from Table3 of [20] and summarized in Table 8. At rst glance, a normalized error of 0.... ..."

### Table 4 Two triangulation-based measures before and after global optimization: rms of 3D reconstruction error in centimeter and normalized stereo camera error

"... In PAGE 10: ... The closer the NSCE to one, the better the calibration accuracy. Table4 shows these two measures computed for 20 images for the neural and polynomial- based approaches using the calibrated parameters before and after the global optimization step. Fig.... In PAGE 10: ... The coplanarity of the squares that actually lie on the same plane, their alignment and the right angles of the squares obviously reflect the quality of the calibration results. One can conclude from Table4 as well as Table 2 that the neural approach is more able to capture the variations of the parameters across the lens settings and it thus provides better calibration results. Moreover, the global optimization step clearly improves the calibration accuracy by tuning and coupling the different parameter formulations.... ..."

### Table 1: Theoretical error in the real-world reconstructions. Each row shows the standard deviation of the error in the 3D points, camera orientations, positions and focal length, the relative amplitude of the residues a2 a26a47a44

2002

"... In PAGE 8: ... For each dataset, the covariance of the estimator is estimated as in [10]. Table1 shows the standard deviation of the error in the 3D points, camera orientation, position and focal length, as well as the amplitude of the residues a2 a26a22a44 a1 a48 a1 a0 a3 a3a2a5a2 relatively to that of the observations a2 . Tour-Eiffel This dataset illustrates how symmetry relations can be used.... ..."

Cited by 2

### Table 1: Theoretical error in the real-world reconstructions. Each row shows the standard deviation of the error in the 3D points, camera orientations, positions and focal length, the relative amplitude of the residues a2 a26a47a44

in Maximum Likelihood 3D Reconstruction From One or More Uncalibrated Views Under Geometric Constraints

2002

"... In PAGE 8: ... For each dataset, the covariance of the estimator is estimated as in [10]. Table1 shows the standard deviation of the error in the 3D points, camera orientation, position and focal length, as well as the amplitude of the residues a2 a26a22a44 a1 a48 a1 a0 a3 a3a2a5a2 relatively to that of the observations a2 . Tour-Eiffel This dataset illustrates how symmetry relations can be used.... ..."

Cited by 2

### Table 1 shows the average 3D reconstruction error along the x,y and z axes when comparing a set of SIFT features with their manually measured correspondences. The result of the experiment indicates that while our vision-based 3D recon- struction algorithm would have an average error of nearly 2cm along the xy plane, depth recon- structions could grow as large as 20cm. These re- sults suggest the use of a measurement covariance matrix for the SLAM implementation with

"... In PAGE 5: ... Table1 . Averrage errors obtained by comparing the manually measured po- sitions and the positions obtained using the vision system.... ..."

### Table 13: Settings for more accurate computations.

"... In PAGE 12: ... In this case, the advantage of the fast multipole method over the standard approach would be even larger. Since the error reduction is not optimal anymore for the finer meshes, the accuracy of the fast multipole method was increase, see Table13 . The following table shows that an increase of DEGREESL and CMP leads to a reduction of the error.... ..."

### Table 9: The results of the computed invariants from Euclidean reconstruction of the conics. The computed invariants are very accurate and stable in both cases.

1995

Cited by 4

### Table 1 Reconstruction error per DOF (in normalized coordinates) Data used 3D reconstruction error

"... In PAGE 12: ...ut the x locations were not. Fig. 7 shows the coordinates used and a sample reconstruction for this experiment. Note that in both cases (see Table1 ), the re- construction is quite accurate in terms of mean- squared error. This shows that the ten learned modes are a su ciently strong characterization to accurately reconstruct the 3D lip shape from 2D data.... ..."