This study proposes an intelligent analysis method for students’ movement performance data by integrating binocular vision technology, BP neural network and subgroup computation. Through dual-camera stereo correction and block matching algorithms, high-precision depth information and motion data are acquired, a threelayer BP neural network model based on correlation analysis to screen key indicators is constructed, the nonlinear mapping ability is optimized by combining Sigmoid function, and data normalization and sample grouping are used to improve the model generalization performance. The performance indicator (PI) clustering algorithm is further introduced to narrow the differences of dependent variables within the cluster, and the model training accuracy and efficiency are significantly improved by base adjustment and scaling. The multidimensional analysis showed that the physiological indicators such as morning pulse, blood pressure and blood oxygenation were significantly and dynamically associated with the athletic performance. Compared with the least squares support vector machine and hybrid genetic neural network, the proposed method shows better prediction stability and real-time performance in eight physical measurements, especially in standing long jump (99.41%) and BMI (99.53%), with an average accuracy of 98.2%, which is more than 10% higher than that of the traditional method, and the single prediction time is only 1.3 seconds, which verifies the combination of real-time performance and accuracy of the proposed method. real-time and accuracy.
