Research on the effectiveness of posture correction for basketball shooting in vocational colleges based on video analysis

To address the low efficiency and high subjective error of manual correction in basketball shooting instruction in vocational colleges, this study constructed a shooting posture correction model based on video analysis and validated its effectiveness. The model integrates high-definition video visual data with wrist/elbow acceleration data and employs a hybrid MobileNet- CNN + Attention-LSTM architecture to achieve real-time extraction of key joint features, automatic error identification, and generation of personalized correction suggestions. Sixty vocational college students with no professional basketball training experience were randomly divided into an experimental group (model correction) and a control group (traditional manual correction) for an eight-week experiment. Results showed that the experimental group's movement standardization score increased from 62.3±5.8 to 89.7±3.2, a 44.0% improvement, significantly higher than the control group's 28.0% (P<0.001). Their shooting accuracy increased by 27.3 percentage points, compared to the control group's 15.7 percentage points (P<0.001). Errors in key joints, such as the shoulder-elbow angle and elbow-wrist angle, decreased by over 60%, compared to the control group's approximately 35% (P<0.001). The model processed a single frame in 28ms and achieved an error recognition accuracy of 92.3%, making it suitable for teaching scenarios. The study demonstrated that the model can effectively improve correction outcomes and provide a practical reference for empowering physical education teaching technology in higher vocational colleges.