Maximizing Accuracy of Fall Detection and Alert Systems Based on 3D Convolutional Neural Network
Seokhyun Hwang, DaeHan Ahn, Homin Park, Taejoon Park · 2017
We present a deep-learning-based approach to maximize the accuracy and reliability of vision-based fall detection and alert systems. The proposed approach utilizes a 3D convolutional neural network (3D-CNN) to analyze the continuous motion data obtained from depth cameras and exploits a data augmentation method to do away with overfitting. Our preliminary evaluation results demonstrate that it achieves the classification accuracy of up to 96.9%.