Real -time Fall Detection and Alert System Using Pose Estimation

One of the most prevalent events in elderly persons is falling and not only it can cause mental and physical injuries but also it can cost too much. Consequently, due to the vital importance of developing an intelligent surveillance system to detect fall events and inform the family or the caregivers, we presented a fast and robust approach that can immediately detect falls and inform the caregivers by SMS so they can provide immediate help and consequently, the amount of injury and treatment costs will be reduced. Our approach includes two networks, pose estimation and MLP classifier. At first, we gathered a dataset that includes landmarks of the body at 500 poses including lying and non-lying positions with different illumination settings and then, MLP net is used to classify these poses. The accuracy and loss reached 92.5% and 0.3 respectively, on the validation dataset. The speed of the process is nearly 3 frames per second on core i7-6500U CPU, hence it can be used in real-time.