Implementation of fall detection system based on 3D skeleton for deep learning technique

In this paper, a fall detection system combined with traditional algorithm with the neural network was proposed. Firstly, we propose a skeleton information extraction algorithm, which transforms depth information into skeleton information and extract the important joints of fall. Secondly, we propose a high robustness deep convolution neural network architecture using pruning method to reduce parameters and calculations in the network. The low parameter and calculation is very adapted to implement on embedded system. The experiment results show the high accuracy and robustness on the popular benchmark dataset NTU RGB+D. We also implement it on NVIDIA Jetson Tx2 platform.