Dual-mode exercise pose recognition using video and pressure sensing Matrix Mat

This thesis project presents an advanced pose recognition system that functions effectively both with and without a vision-based system by employing piezoresistive materials. Designed to accurately identify various human poses such as standing, cross-legged sitting, and one-leg standing, this system integrates real-time predictions using either camera inputs by applying pose estimation method or an 8x8 pressure sensing matrix mat with Convolutional Neural Network (CNN). We have enhanced the robustness of the system through the implementation of centering and scaling techniques in the pose-estimation model. Furthermore, by applying Principal Component Analysis (PCA), we have successfully reduced the computational complexity of the data points by half while achieving the desired accuracy for the model. Low burden pressure sensing matrix mat was developed for the system as well. This research holds significant potential for applications in diverse fields including domestic exercise routines, sports performance analysis, physical rehabilitation, and interactive fitness technologies, all with a focus on safeguarding user privacy