Validating a Low-Cost Depth Sensor for Facial Anthropometry: Capabilities and Appropriate Use

Area of Science:

• Biomedical Engineering
• Computer Vision
• Anthropometry

Background:

• Low-cost depth sensors are emerging for 3D facial capture.
• High intraclass correlation coefficients (ICC) are often presented as proof of their readiness.
• However, ICC measures relative agreement, not absolute accuracy.

Purpose of the Study:

• To evaluate the accuracy and precision of a low-cost depth sensor for 3D facial measurements.
• To compare its performance against a structured-light scanner and direct calipers.
• To determine if high ICC values correlate with low absolute error.

Main Methods:

• 12 adults were measured by 2 observers at 16 facial distances.
• Evaluated device-versus-caliper ICC(3,k), Bland-Altman statistics, observer reliability, measurement error, scan-rescan precision, and mean absolute error (MAE).
• Compared against a structured-light scanner and direct calipers.

Main Results:

• Device-versus-caliper ICC was high (0.88 and 0.93).
• Depth-sensor MAE (2.75 mm) was significantly higher than calipers (1.64 mm).
• ICC did not predict absolute error (rho=-0.17). Scan-rescan repeatability was poorer (0.84 vs 0.36 mm).

Conclusions:

• The depth sensor offers cost-effective relative ranking but lacks precision for high-fidelity reconstruction.
• Absolute error, not just ICC, is crucial for assessing clinical usability.
• Suitable for morphometric surveys and within-subject monitoring, but not precise facial capture.