Real-time Pilates Posture Recognition System Using Deep Learning Model

Abstract As the pandemic situation continues, many people exercise at home. Mat Pilates is a popular workout and effective core strengthening. Although researchers have conducted pose recognition studies for posture correction, study on only one case static images. Therefore, purpose of monitoring, we propose real-time system smartphone monitoring. We aimed to recognize 8 exercises—Bridge, Head roll-up, Hundred, Roll-up, Teaser, Plank, Thigh stretch, Swan. First, Blazepose model used extract body joint features. Then, designed deep neural network that recognizes based extracted It also measures number workouts, duration, similarity experts in video sequences. The precision, recall, f1-score are 0.90, 0.87, 0.88, respectively. introduced application expected be management