A framework for elders fall detection using deep learning

Elders are one of the huge and fast-growing populations, the majority of them suffers from several daily accidents like fall, and according to the World Health Organization, fall is the second leading cause of accidental injury deaths, for this reason, we proposed a dynamic video classification system in order detect fall with one surveillance camera. Our proposed framework consists of two steps, the first step is the extraction of the human body silhouette from the video frames with the Mask R-CNN, in the second step, we used a combination between the Convolutions Neural Network CNN, and the LSTM long short-time memory to learn long-term dependencies between the successive frames. The best result of our system was achieved when we used 10 frames per video with accuracy, precision, recall, and F1-score equals to 100%.