(1) Background: Gait analysis provides quantitative information on walking patterns and has proven invaluable for assessing motor function in rehabilitation programmes. A markerless motion capture system combining computer vision techniques provides low-cost, real-time, portable gait analysis. (2) Methods: The kinematics of the knee and ankle of twenty-seven healthy volunteers were assessed using a single smartphone camera combined with the MediaPipe human pose estimation framework. The system was validated using the OPAL wearable sensor system by APDM Wearable Technologies. (3) Results: Findings showed close correspondence between the two systems for knee kinematics showing a mean absolute error of 4.10° ± 2.32° and 3.15° ± 3.10° for right and left knee flexion, respectively, and a mean absolute error of 2.30° ± 2.01° and 3.12° ± 2.63° for right and left knee extension, respectively. The mean absolute error for right and left knee range of motion was found to be 4.55° ± 3.12° and 4.15° ± 3.01°, respectively. Moreover, Bland–Altman plots indicated minimal bias (average 0.6 for flexion, average 0.47 for the extension, and 0.30 for the range of motion) and excellent correlation for knee flexion bilaterally (0.916 and 0.845 for the right and left side, respectively), with slightly lower but still satisfactory agreement for knee extension (0.862 and 0.845 for the right and left side, respectively). Conversely, ankle measurements revealed poor concordance: dorsiflexion and range of motion presented significant differences and systematic errors, while plantarflexion showed no statistical difference but weak correlation. (4) Conclusions: This study demonstrated that combining a smartphone camera with a human pose estimation framework allows for low-cost, real-time, portable gait analysis, particularly of the knee joint.
Russo et al. (2026) studied this question.