Skeleton Based Fall Detection with Convolutional Neural Network

There are more and more elderly people living alone in the society today, and the supervision of the elderly has become a social problem. In order to enable elderly people living alone to be found at home when they fall, minimize the damage to them. We propose a skeleton-based real-time fall detection system. Firstly, we encode the sequence of motions into an RGB image and retain the original spatial structure information and temporal dynamic information. Then, we designed a lightweight convolutional neural network (CNN) to recognize the encoded image. We collected a fall dataset and merged it with the MSRDaily Activity3D dataset. We use two data enhancement methods to extend the dataset. The results of our model on the test set show that the recall is 93.9% and the accuracy is 93.75%. Finally, we also propose an online detection algorithm. The experimental results show that the recall is 1, and the accuracy is 95%.