Sixty patients with knee osteoarthritis were randomized to receive either a computer vision–based graded exercise application (n=32) or exercise education videos (n=28). The primary outcome was change in pain, physical function, and stiffness measured by the WOMAC index. Secondary outcomes included self-efficacy, affective state, quality of life, range of motion, and user experience. Outcomes were assessed at baseline and 6 weeks. Overall, the results of the study revealed that the application significantly improved physical function and self-efficacy compared with education videos, though no significant differences were seen in pain or stiffness. The findings suggest that a computer vision–based graded exercise program enhances functional outcomes and self-efficacy, making it a promising tool for digital rehabilitation.