Real Time Fall Detection Model with Action Recognition based on Deep Learning

Healthcare practitioners are making great use of the benefits of most cutting-edge technologies, which is improving the area of healthcare on a scalable basis. The shift from manual to very accurate virtual monitoring has occurred. An important global concern is the care of the elderly. A fall is among the worst things that may happen to an aged person. With the ageing population expanding faster than ever, fall detection systems are essential. Fall detection and rescue systems may now quickly offer emergency assistance to victims of unintentional falls, thanks to advancements in technology. This makes it possible to provide appropriate care and lower the number of fatalities brought on by falls. We present a deep convolutional neural network based real-time fall detection model. Because it provides real-time monitoring and fall detection, an automated fall detection model is essential for elder care. Manual fall detection and timely intervention depend on several parameters being available in a timely manner. Real-time monitoring systems and deep learning approaches can be used to do this. Many deep learning networks can be used for fall detection. We present an effective approach using torch tensors and deep convolutional neural networks for real-time fall detection. From a collection of real-time input videos, the system will distinguish between fall and nonfall and respond accordingly. The user's actions and the fall will be recognized by the system. An automated fall detection system in hardware was demonstrated using an LSTM neural network model on an Up Extreme Series single-board computer. A summary of the experiments can be found in the appendix.