Assessing the precision of 3D human model for forensic biomechanics application

Three-dimensional (3D) human modeling is increasingly used in forensic biomechanics to reconstruct spatial relationships within incident scenes. A critical requirement is anthropometric fidelity. This study quantified the agreement between a customizable 3D human model and measured human anthropometry. Eight key anthropometric dimensions, including stature, reach distances, limb lengths, and span, were collected from each participant and their corresponding stature-scaled virtual 3D models generated using the Human Generator add-on for Blender. Errors were expressed as mean absolute percentage error (MAPE) and mean signed percentage error (MPE), with two-sided 95% confidence intervals estimated by percentile bootstrap (10,000 resamples). Agreement was high for stature (MAPE <1%, MPE ≈ 0 for both sexes). Several segment and reach measures showed small errors around 5% or under. Two systematic discrepancies were identified. First, SPAN was consistently underestimated with a bias in males of approximately -8% and a MAPE of approximately 13.5%, and MPE of -3.9% and MAPE of 5.7% with females. Second, LOWER LEG was overestimated in males exclusively (MPE ≈ +8.7% and MAPE ≈ 8.7%). These findings highlight both the capability of such models for forensic applications while also highlighting some of the potential pitfalls. While certain measurements exhibit strong alignment, forensic practitioners must be aware of systematic errors when using stature-scaled models.