Accuracy Verification and Correction of a Depth-Camera-Based Digital Measurement System for Pelvic Balance Assessment in the Trendelenburg Test

This study proposes a markerless, single-camera depth-sensing system for digitally quantifying pelvic-balance ability in the Trendelenburg test, and verifies its accuracy and reliability against a marker-based motion capture (MC) reference and manual measurement (MM) in 16 healthy adults. A paired-samples t-test confirmed that the pre-correction bias between MC and DC was significantly different from zero (p < 0.001), whereas the post-correction bias between MC and the corrected DC (cDC) was not (p = 0.791; LOSO p = 0.771), directly confirming that a correction coefficient (a = 89.42%) derived from this systematic bias successfully removed it: Bland–Altman analysis showed that the mean bias (−5.11 mm) was reduced to near zero after correction (0.18 mm), and Lin’s concordance correlation coefficient (CCC) improved correspondingly from 0.944 to 0.981. These findings were consistent with intraclass correlation coefficient (ICC) analysis, which showed excellent consistency (0.980) and absolute agreement (0.946) with MC even before correction, further improving to 0.982 after correction. All improvements were confirmed using Leave-One-Subject-Out (LOSO) cross-validation (CCC = 0.980; absolute-agreement ICC = 0.981; bias = 0.21 mm), indicating that they are not an artifact of evaluating the correction on the same data used to derive it. By contrast, MM showed poor reliability and agreement with MC (CCC = 0.258; ICC = 0.264–0.440), underscoring the limitations of visual, examiner-based assessment. These findings indicate that the proposed single-camera depth-sensing system offers a convenient, markerless alternative to marker-based motion capture, with accuracy and reliability suitable for practical deployment, and that the verification–correction methodology provides a generalizable approach for improving the accuracy of similar low-cost digital measurement systems.