Three-dimensional facial capture using a custom-built photogrammetry setup: Design, performance, and cost

Abstract

Introduction: Although stereo-photogrammetry is increasingly popular for scanning faces three-dimensionally, commercial solutions remain quite expensive, limiting their accessibility. We propose a more affordable, custom-built photogrammetry setup (SF3D), and evaluate its variability within-and between-systems. Methods: 29 subjects and a mannequin head were imaged three times using SF3D and a commercially available system (3dMDFace). Next, an anthropometric mask was mapped visco-elastically onto the reconstructed meshes using Meshmonk. Withinsystems shape variability was determined by calculating the RMSE of the Procrustes distance between each of the three subject's scans and the subject's ground truth (calculated by averaging the mappings after a non-scaled generalized Procrustes superimposition). Inter-system variability was determined by similarly comparing the ground truth mappings of both systems. Two-factor Procrustes ANOVA was used to partition the inter-system shape variability to better understand the source of the discrepancies between the facial shapes acquired by both systems. Results: The RMSE of the within-system shape variability for 3dMDFace and SF3F were 0.52 +/-0.07 mm, and 0.45 +/-0.17 mm, respectively. The corresponding values for the mannequin head were 0.42 +/-0.02 mm, and 0.29 +/-0.03 mm, respectively. The between-systems RMSE was 1.6 +/-0.34 mm for the study group, and 1.38 mm for the mannequin head. Two-factor ANOVA indicated that variability attributable to the system was expressed mainly at the upper eyelids, nasal tip and alae, and chin. Conclusions: The variability values of the custom-built setup presented here were competitive to a state-of-the-art commercial system at a more affordable level of investment.

Accuracy and Reliability of Kinect Motion Sensing Input Device’s 3D Models: A Comparison to Direct Anthropometry and 2D Photogrammetry

“…The present study found agreement between 2D photogrammetry and other methods, which suggests that 2D measurements can be safely used in images collected with accurate technique and attention. Farkas et al [29] and Wellens et al [30] reported that, by the distortions in photogrammetry, the difference between direct anthropometry and 2D photogrammetry may be caused. It seems crucial to make the calibration in the same plane as the measuring points in 2D measurements.…”

Integration and Application of Multimodal Measurement Techniques: Relevance of Photogrammetry to Orthodontics

“…Other depth-acquisition techniques include: depth from defocus, depth from focus, pulse ranging, time-of-flight, phase-based, or continuous wave ranging methods [ 1 ], in addition to confocal, active wave sampling [ 2 ]. Types of 3D scanners relevant for orthodontic applications include (this list is by no means exhaustive): for face scanning: Vectra, 3DiD, 3dMD [ 3 ], EinScan Pro and Planmeca [ 4 ], Konica Minolta 910, and Polhemus [ 5 ]; desktop scanners for scanning plaster dental models: Konica Minolta VI-9i and Imetric laser scanner [ 6 ]; intraoral scanners: Trios, iTero Element, Dental Wings, Panda 2, Medit i500, Planmeca, and Aoralscan [ 7 ]. …”

“…for face scanning: Vectra, 3DiD, 3dMD [ 3 ], EinScan Pro and Planmeca [ 4 ], Konica Minolta 910, and Polhemus [ 5 ];…”

“…According to a review article covering a very wide collection of studies analysing the accuracy of digital imaging of facial, skeletal, and intraoral tissues [ 31 ], the measured trueness for face scanners ranges from 0.140–1.380 mm and, for intraoral scanners, from 0.017–0.378 mm. In particular, the accuracy of face-mask scanning was about 0.10 mm [ 5 ] and of mannequin face scanning 0.52 mm [ 3 ]; for the 3dMD face scanner, the accuracy was 0.067 mm [ 32 ], while for the Konica Minolta Vivid 910 scanner, the accuracy was 0.08 mm [ 5 ]. Intraoral scanning achieves remarkable accuracy over small areas [ 7 , 33 ].…”

“…Some current face scanners use photographs of the patient’s face acquired from several viewpoints, obtained at the same moment in time. However, their cameras are placed either on an arm, a rig, or a frame [ 3 ]. This ensures their geometric relationships remain constant, and their earlier calibration is possible, which greatly simplifies computation and improves accuracy.…”