A real-time fall detection of elderly people in indoor environments

The extraordinary effectiveness of existing image-based AI technologies and the resulting interest in their use in key decision-making processes have sparked a rise in attempts to make such intelligent systems visible and understandable. The technology we use today is the result of several researchers achieving artificial intelligence achievements. Artificial intelligence leverages computers and machines to mimic the human mind's problem-solving and decision-making capabilities. For example, some researchers have demonstrated that convolutional neural networks with shorter connections between layers adjacent to the input and those close to the output can be significantly deeper, more accurate, and efficient to train. This study serves as a framework for distinguishing fall event from other indoor natural human activities. In this study, a modified DenseNet 121 approach that relies on the deletion of some layers and the addition of others for fall detection is presented. With the proposed approach, losses are reduced, and network variables like weights and learning rate are updated using the binary cross-entropy loss function. Finally, the sigmoid classifier identifies human falls using binary classification. It is shown that the proposed system achieves a precision of 98.83% on experiments, outperforming existing modern models. Key Words: Convolutional Neural Networks, Deep Learning, Fall Detection.