Direction-Independent Graph-Based GCN for Fall Detection

As the global population ages, the issue of falls among the elderly has become an increasingly pressing concern, garnering significant attention from the medical and research communities. Among the plethora of fall detection methods, inertial measurement units (IMUs) are favored by many researchers and practitioners for their capacity to capture real-time dynamic motion changes. However, the irregular change of direction during the fall process increases the difficulty of fusing and extracting features from the accelerometer and gyroscope data that are originally sensitive to the original direction. In response to this challenge, this study introduces the Direction-Independent Graph-Based Convolutional Neural Network (D-GCN). The D-GCN integrates features from acceleration and angular velocity data to form a novel topological representation. To assess the efficacy of D-GCN in fall detection, a dataset comprising 16 types of human movements, including 12 resembling falls and 4 actual falls, was established. The experiments demonstrate that this method achieves an accuracy of 0.901 on the dataset, which is an improvement of 11.9% over traditional machine learning algorithms.