Research on Prediction Technology of Table Tennis Landing Point Based on Multi-scale Image Processing

This study is devoted to improving the accuracy and robustness of table tennis impact prediction through multi-scale image processing technology. By using high-resolution camera to capture the scene of table tennis match, a multi-scale image processing model based on Gaussian pyramid is constructed in this study, aiming at capturing the characteristics of table tennis more comprehensively and carefully. During the experiment, firstly, the image is Gaussian filtered and downsampled to construct a multi-layer pyramid, so as to obtain the image representation at different scales. Then, local features at different scales are extracted by local binary pattern (LBP) and directional gradient histogram (HOG) to form multi-scale feature representation. An end-to-end prediction model of table tennis impact point is constructed by fusing features at different scales. Comparative experiments show that the model based on multi-scale image processing shows better performance in different scenes and lighting conditions compared with traditional methods. The accuracy of the model is significantly improved, especially in the situation of dealing with high-speed movement and complex spatial structure. It is also found that the reasonable adjustment of the scale selection plays a key role in the performance of the model, so that it can achieve excellent prediction results in different table tennis stadiums. The results of this research not only have a positive application prospect in the field of table tennis, but also have certain guiding significance for the development of image processing and prediction technology in sports.