Video surveillance of elderly person aimed to healthcare via deep learning

The increasing concentration of the human population living in cities increases the demand for smart city services. Among these services, an important sub-title under the title of smart health is the elderly care service. Today, the life expectancy of people is increasing and at the same time, there is an increase in diseases such as Alzheimer's that cause need for care. Fall detection and monitoring the activities of these people while living alone in the hospital room or home are among the research topics that have been studied extensively under the title of smart healthcare. Because, it is very important to warn the authorities as soon as a fall occurs. IoT technology, which makes a great contribution to the development and spread of smart city services, creates big data. In big data analytics, a very effective artificial intelligence (AI) solution is deep learning networks. In this study, an original dataset was created from videos taken representing people living in a room to train a deep learning network in the Convolutional Neural Network (CNN) model. Using this data set, CNN models, which are designed in small size and have different hyperparameters, are trained via Matlab software, as they are intended to work in real time. The training process was evaluated with performance criteria based on the confusion matrix. With these trained networks, activity monitoring and fall detection are possible in real-time video captured by a webcam. The time distributions of the activities according to standing, sitting, sleeping, bending, falling and empty room situations were plotted. Thus, it will be possible to notify as soon as a fall occurs and to share activity reports with the authorities at periodic intervals via cloud computing. Efficiency evaluation of trained networks owning different structures was made. In addition to these classifications based on photo frames obtained from the video stream, it has been seen that it is necessary to distinguish between intended sitting and involuntary falling events by adding a logical sorting process or Long Short-Term Memory (LSTM) classifier.