Fall recognition using a three stream spatio temporal GCN model with adaptive feature aggregation

The prevention of falls is paramount in modern healthcare, particularly for the elderly, as falls can lead to severe injuries or even fatalities. Additionally, the growing incidence of falls among the elderly, coupled with the urgent need to prevent suicide attempts resulting from medication overdose, underscores the critical importance of accurate and efficient methods of detecting a fall. This makes a computer-aided fall detection system necessary to save elderly people's lives worldwide. Many researchers have been working to develop fall detection systems. However, the existing systems often struggle with problems such as unsatisfactory accuracy, limited robustness, high computational complexity, and sensitivity to environmental factors. In response to these challenges, this paper proposes a novel three-stream spatio-temporal feature-based human fall detection system. Our system incorporates joint skeleton-based spatial and temporal Graph Convolutional Network (GCN) features, joint motion-based spatial and temporal GCN features, and residual connections-based features. Each stream employs adaptive graph-based feature aggregation and consecutive separable convolutional neural networks (Sep-TCN), significantly reducing the computational complexity and the number of parameters of the model compared to prior systems. Experimental results on multiple datasets demonstrate the superior effectiveness and efficiency of our proposed system, with accuracies of 99.68%, 99.97%, 99.47 % and 98.97% achieved on the ImViA, Fall-UP, FU-Kinect and UR-Fall datasets, respectively. The remarkable performance of our system highlights its superiority, efficiency, and generalizability in real-world human fall detection scenarios, offering significant advancements in healthcare and societal well-being.