Accelerometer-Based Human Fall Detection Using Convolutional Neural Networks.

Area of Science:

• Computer Science
• Artificial Intelligence
• Health Informatics

Background:

• Human falls represent a significant global public health concern, leading to millions of injuries and deaths annually.
• The economic impact of falls includes direct healthcare costs and indirect losses in societal productivity.
• Effective human fall detection and prevention are critical areas of health research.

Purpose of the Study:

• To propose a deep learning model for human fall detection within an Internet of Things (IoT) and fog computing framework.
• To address challenges related to dimensionality and model simplicity in fall detection systems.
• To evaluate the performance of the proposed model using established metrics and datasets.

Main Methods:

• Development of a Convolutional Neural Network (CNN) comprising three convolutional layers, two max-pooling layers, and three fully-connected layers.
• Utilizing three open datasets for model evaluation and comparison with existing research.
• Implementing data augmentation techniques during the training phase to improve model robustness.

Main Results:

• The proposed CNN model demonstrated strong performance in human fall detection.
• Data augmentation significantly enhanced the model's accuracy, precision, sensitivity, specificity, and Matthews Correlation Coefficient.
• The approach effectively managed dimensionality and modeling simplicity issues.

Conclusions:

• Deep learning, specifically CNNs, offers a promising solution for accurate human fall detection in IoT environments.
• Data augmentation is a crucial technique for optimizing the performance of fall detection models.
• Further research is needed to address ongoing challenges and explore future directions in fall detection technology.