Purpose <p>This study aimed to&#xa0;develop a dual-camera markerless system based on data from patients with osteoarthritis (OA) and validate its agreement with a marker-based motion capture system for measuring lower-limb kinematics.</p> Methods <p>A total of 152 OA patients were divided into a training set (<i>n</i> = 120) and a test set (<i>n</i> = 32). Kinematic data during gait were collected simultaneously via both markerless and marker-based systems. The dual-camera markerless system consists of a 2D pose extractor based on a neural network and 3D triangulation, and the kinematic differences between the two systems were evaluated via the&#xa0;root mean square distance (RMSD) and root mean square error (RMSE) and intraclass correlation coefficient (ICC).</p> Results <p>The markerless system demonstrated great performance, achieving a grand mean RMSD of 11.0&#xa0;mm and an ICC of 0.95 for keypoints. Joint angle analysis revealed a mean RMSE of 4.25°, with ICC values for joint angle waveforms reaching 0.90 in the sagittal plane, 0.48 in the frontal plane, and 0.24 in the transverse plane compared with the marker-based system.</p> Conclusion <p>These results indicate that the dual-camera markerless system provides accurate lower-limb kinematic measurements for patient populations while offering significant advantages in terms of cost-effectiveness, installation simplicity, and reduced operational expertise requirements, facilitating efficient biomechanical assessment in clinical use.</p>