Human fall direction recognition in the indoor and outdoor environment using multi self-attention RBnet deep...

Area of Science:

• Gerontology
• Computer Science
• Artificial Intelligence

Background:

• Falls pose a significant health risk to the elderly, increasing with global population growth.
• Effective fall detection is crucial for timely medical intervention and supporting independent living for older adults.
• Existing safety monitoring systems require advancement for improved accuracy and efficiency.

Purpose of the Study:

• To propose a novel deep learning architecture for human fall direction recognition.
• To develop and evaluate advanced residual block and self-attention mechanisms for fall detection.
• To optimize feature selection and classification performance using a tree seed algorithm.

Main Methods:

• Developed four novel residual block-deep convolutional neural network (RBNet) self-attention models (3-RBNet, 5-RBNet, 7-RBNet, 9-RBNet).
• Extracted deep features from self-attention layers after training models on enhanced images.
• Applied the tree seed algorithm to optimize features from 7-RBNet and 9-RBNet models for improved classification and reduced computational cost.

Main Results:

• The 7-RBNet and 9-RBNet self-attention models demonstrated superior accuracy and precision.
• The proposed method, utilizing the tree seed algorithm on 7-RBNet and 9-RBNet features, achieved maximum accuracies of 93.2% and 92.5% on a human fall dataset.
• The approach showed improved accuracy and precision compared to recent fall detection techniques.

Conclusions:

• The novel deep learning architecture and optimization algorithm effectively recognize human fall direction.
• The 7-RBNet and 9-RBNet self-attention models, optimized with the tree seed algorithm, offer a highly accurate and precise solution for elderly fall detection.
• This research contributes to enhanced safety monitoring for older adults, particularly those living independently.