Actions of basketball players integrating computer vision and sports biomechanics

Abstract

To address the issues of subjectivity, data acquisition challenges, low analysis efficiency, and lack of real-time feedback in traditional basketball player action research, this study applied a DeepLabCut model that integrates computer vision and sports biomechanics. By combining image processing and keypoint detection techniques, the model effectively recognizes and analyzes basketball player actions, offering a significant improvement over conventional methods. Firstly, the collected action videos were subjected to data preprocessing, including image size adjustment, grayscale processing, and normalization; then, the DeepLabCut model was utilized for keypoint detection to identify the key action parts of basketball players; finally, the detection results were analyzed and validated through the calculation of biomechanical parameters in sports. The research results showed that DeepLabCut performed excellently in keypoint detection accuracy and processing speed. Compared to traditional methods, its average error was only 4.63 pixels and the average processing time was only 45.7 milliseconds. This research highlights the significance of DeepLabCut and sports biomechanics in enhancing the accuracy and efficiency of action analysis in basketball player action research.