Effects of Computer Vision Technology on Learners’ Performance in Physical Education: A Three-Level Meta-Analysis

In physical education (PE), computer vision (CV) technologies are increasingly used to analyze learners’ movements and provide performance-related feedback. However, the overall effectiveness of CV-assisted instruction and the conditions associated with variations in its effects remain unclear. This meta-analysis examined the effects of CV-assisted instruction on learners’ sports performance. A systematic search identified 20 eligible studies involving 1789 participants and 67 effect sizes. A three-level random-effects model was applied to account for the dependency of multiple effect sizes within the same study. The results showed a significant positive effect of CV-assisted instruction on sports performance compared with comparison conditions (Hedges’ g = 1.37, 95% CI [0.92, 1.83], p < 0.001). However, substantial heterogeneity was observed at both the between-study and within-study levels. A study-clustered CR2 robust variance analysis produced a similar pooled estimate, indicating that the overall effect was robust to within-study dependence. Exploratory moderator analyses showed significant differences according to CV type, learners’ background, and score type. Larger effects were observed for 3D than 2D CV and for experienced than inexperienced learners. Process-oriented measures of movement quality also produced larger effects than outcome-oriented performance measures. In contrast, teaching type, sport type, personalization of CV technology, and intervention duration were not significant moderators. Leave-one-study-out analyses showed that the pooled effect did not depend on any single study. A sensitivity analysis restricted to studies with low overall risk of bias also produced a significant positive effect. Overall, CV-assisted instruction appears to support learners’ sports performance in PE, although the substantial heterogeneity suggests that its effects vary across studies and instructional contexts. These findings indicate that CV type, learner experience, and outcome measurement should be considered when designing and evaluating CV-supported PE interventions. PROSPERO Register: CRD420261422540.