Personalized clothing recommendation framework based on the fusion of sports biomechanics and computer vision

This research aims to create a novel framework that merges sports biomechanics and computer vision for automating the clothing suggesting process, with advancements in extracting biomechanical features, undertaking visual analysis, performing multimodal data fusion, and personalization modeling. The framework employs powerful computer vision techniques and deep neural networks alongside biomechanical sensors like the goniometer, pressure scanner, and other sensors capturing locomotor dynamics. In this study, for the first time, a profound fusion between multidimensional biomechanical variables and captured appealing semantic and visual components is made, with quantifiable relations between the functionality and aesthetic performance of the clothing design established. Judith, the core autonomous system, achieves high-accuracy personalized recommendations through analysis of joint movements, recognition of motion habits, and modeling of pressure distribution. In the framework, an entirely new paradigm for the clothing market is constructed by combining real and virtual models. The system solves the cold-start issue by utilizing cyclic domain transfer learning together with biomechanical features-driven analysis. The obtained results are impressive, with the system achieving a recall of 0.845, precision of 0.892, and NDCG of 0.901, as well as biomechanical-special metrics of body-fit score equal to 0.885, motion comfort 0.873, and pressure distribution uniformity 0.891. As different user groups were analyzed, the results were unchanged. This shows the framework’s practical usability and sustainability. Besides, it opened up a new avenue for intelligent recommendation systems that integrate biomechanical analysis.