Precise Recognition and Evaluation of Men’s Long Jump Based on Transformer and 3D Convolution

For the long jump, the accuracy of action recognition and evaluation affects subsequent training and development plans. Current recognition and evaluation models lack specificity for men’s long jump actions and have low efficiency, which hinders precise and personalized technical improvement and athletic potential development. To address these issues, this study raises a model for precise recognition and evaluation of men’s long jump based on Transformer and 3D Convolutional Neural Network. The model first extracts local action features through a 3D convolutional neural network and then extracts global features using Transformer to obtain relevant action recognition results. On this basis, the study matches the movement posture of feature images with that of standard images according to the characteristics of human skeletal motion to complete the quality evaluation of men’s long jump actions. Experimental results show that in terms of algorithm performance, the area under the curve at the recognition level reaches 0.927, which is significantly higher than that of comparison algorithms. The Coefficient of Determination for evaluation reaches 0.912, demonstrating good algorithm performance. In practical applications, the model’s maximum memory usage is 315.4 MB, and when the data volume is 200, its maximum response time is 48.2 s. Furthermore, the model achieves the highest accuracy in actual men’s long jump recognition and evaluation tasks. This model provides new directions and insights for the fields of action recognition.