Validation of AI-Based Markerless Gait Event Detection during Perturbed Walking Using Smartphone Videos from Two Camera Perspectives

Background: Perturbation-based balance training (PBT) requires objective quantification of reactive stepping responses, yet step detection during perturbed walking remains methodologically challenging. Markerless, smartphone-based gait analysis offers a scalable alternative to laboratory motion capture systems, but its robustness under perturbation-induced gait irregularities is insufficiently validated. This pilot study aimed to validate the SMARTGAIT markerless motion analysis system against frame-accurate manual annotations during normal and perturbed treadmill walking in older adults.Methods: Five geriatric participants (mean age 83 ± 4.47 years; 40% female) walked on a perturbation treadmill while eight unexpected perturbations in both mediolateral and anteroposterior displacement were administered. Smartphone video from frontal and diagonal perspectives were recorded and gait events (initial contact, final contact) were manually annotated as ground truth. Model performance was evaluated using F1 scores with a ±33 ms tolerance window. Retraining was conducted using five-fold cross-validation based exclusively on perturbed segments.Results: Across 2477 annotated steps (2106 normal; 371 perturbed), initial detection performance was consistently lower during perturbed walking (median F1 score: frontal 0.49; diagonal 0.88) compared to normal walking (median F1 score: frontal 0.68; diagonal 0.96), with large effect sizes despite non-significant p-values (0.06–0.10). No statistically significant differences in step detection performance were observed between diagonal and frontal camera perspectives for normal walking (p = 0.125) or perturbed walking (p = 0.063); however, large effect sizes (normal walking: r = 0.867 and perturbed walking: r= −1.000) suggest that camera perspective may influence detection accuracy depending on walking condition. Retraining improved perturbed gait event detection in the frontal perspective (0.49 → 0.76; median difference 0.30, 95% CI [0.08–0.47]) but showed minimal change in the diagonal perspective.Conclusion: The smartphone-based markerless motion analysis approach SMARTGAIT demonstrated high validity for normal walking and moderate robustness under perturbed conditions. Camera perspective did not substantially influence performance, but targeted retraining partially mitigated performance degradation under perturbations. Larger-scale validation studies are required before clinical implementation.