Integrated Frameworks for Three-Dimensional Scene Reconstruction - 3D Computer Vision: Efficient Methods and Applications - page 222

Ta b l e 5 . 4 Three-dimensional reconstruction results for the synthetic ground truth surface shown
in Fig. 5.16 a. If not identified otherwise, the results were obtained with the global optimisation
approach. The error values for z are given in pixels
Method, utilised information Figure RMSE (without noise) RMSE (with noise)

z p q z p q

I 1 5.16 d 1.11 0.046 0.077 1.11 0.047 0.077

φ 1 - 2.13 0.102 0.059 3.92 0.163 0.117

I 1 , I 2 5.16 e 0.22 0.012 0.018 0.21 0.014 0.019

I 1 , I 2 (local) 5.16 j 0.00 0.000 0.000 0.19 0.046 0.088

I 1 , φ 1 5.16 f 0.17 0.012 0.007 0.19 0.040 0.065

I 1 , φ 1 (local) 5.16 k 0.00 0.000 0.000 0.28 0.134 0.210

I 1 , D 1 - 1.11 0.044 0.077 1.13 0.098 0.102

I 1 , D 1 (local) - 2.12 0.088 0.178 7.17 0.837 0.928

I 1 , φ 1 , D 1 - 0.01 0.001 0.001 0.52 0.103 0.088

I 1 , φ 1 , D 1 (local) - 0.00 0.000 0.000 0.31 0.149 0.214

I 1 , I 2 , φ 1 - 0.11 0.009 0.005 0.20 0.034 0.066

I 1 , I 2 , φ 1 (local) - 0.00 0.000 0.000 0.35 0.074 0.099

I 1 , I 2 , φ 1 , φ 2 - 0.01 0.001 0.001 0.21 0.056 0.078

I 1 , I 2 , φ 1 , φ 2 (local) - 0.00 0.000 0.000 0.24 0.057 0.079

z 5.16 g 0.14 0.012 0.013 0.20 0.034 0.036

I 1 , z 5.16 h 0.11 0.008 0.009 0.15 0.023 0.028

I 1 , φ 1 , z 5.16 i 0.09 0.006 0.005 0.13 0.042 0.065

I 1 , I 2 , φ 1 , z - 0.09 0.006 0.005 0.15 0.036 0.062

I 1 , I 2 , φ 1 , φ 2 , z - 0.07 0.004 0.003 0.11 0.052 0.078

Fig. 5.17 Dependence of the reconstruction error of the SfPR approach on the weight parameters
λ and μ according to ( 5.21 )and( 5.32 ) for the synthetic example shown in Fig. 5.16 a without noise
( left ) and with noise ( right )

the reconstruction result becomes very accurate. Similarly accurate reconstruction
results, however, are already obtained based on a single intensity and polarisation
image.