Convolutional Neural Network-Based Fall Detection for the Elderly Person Monitoring

The purpose of this paper is to present a generalized human fall detection system for elderly assistance. The human population above the age of 60 is constantly growing. India is the world's second-most peopled country, having 76.6 million individuals aged 60 and above, occupying more than 7.7% of the total population. Falls are considered a significant problem among the elderly. Falls are a significant source of mortality among the elderly. As a result, instant medical attention is required following a fall. When related to wearable sensors, vision-based fall detection is a more appropriate scheme for supervising old persons. Since a variety of backgrounds and scenes are available, fall event detection necessitates an intelligent approach for extracting the relevant feature. Deep learning algorithms have demonstrated very excellent classification performance in recent years. Compared to traditional techniques, fall detection systems using Convolution Neural Networks (CNN) are highly competent in detecting fall occurrences. The value of the proposed research lies in the CNN-based Fall Detection (FD) system which is data-independent. CNN is infamous for being difficult to tune and data-intensive. There is a risk of overfitting with only a few constructive cases of anomalies among hours of footage. The value of the proposed system lies in combination of diverse datasets from various human fall scenes of office, home, coffee room and lecture room for extraction of novel feature sets to be input to CNN model. The proposed system addresses vital issues the healthcare underthought by automated human fall detection with an accuracy of 93.81% by combining the FDD, MCFD, SDU, and URFD.