Abstract
We present a novel local-global feature fusion framework for body-weightexercise recognition with floor-based dynamic pressure maps. One step furtherfrom the existing studies using deep neural networks mainly focusing on globalfeature extraction, the proposed framework aims to combine local and globalfeatures using image processing techniques and the YOLO object detection tolocalize pressure profiles from different body parts and consider physicalconstraints. The proposed local feature extraction method generates two sets ofhigh-level local features consisting of cropped pressure mapping and numericalfeatures such as angular orientation, location on the mat, and pressure area.In addition, we adopt a knowledge distillation for regularization to preservethe knowledge of the global feature extraction and improve the performance ofthe exercise recognition. Our experimental results demonstrate a notable 11percent improvement in F1 score for exercise recognition while preservinglabel-specific features.