A Study on Fusion Acquisition of Image and CSI Data for Posture Estimation in Indoor Environments

This research suggests an innovative approach to acquiring and synchronizing image-based data and non-image-based data (CSI) for posture estimation in indoor environments. The method proposed involves using Python's Multiprocessing library to simultaneously gather CSI data and video data in parallel. When a person is detected using the YOLOv8-pose algorithm, CSI data is collected every 0.1 seconds, and video frames are saved with timestamps. The synchronization process combines the timestamps of both datasets using linear interpolation. The Bi-LSTM model, trained with the synchronized data, showed high accuracy in human activity recognition (HAR). This study's results highlight the potential for efficient and reliable HAR through data fusion, with applications in smart homes, crime prevention, and sports analysis.