A DEEP LEARNING MODEL FOR FALLING DETECTION

Abstract

Falling is one of the most serious problems for humans, accounting for up to 25% of death rates, which is even higher for the elderly. Falling detection is one of the most important problems in computer vision. In recent years, computer vision has made impressive progress when deep learning demonstrates the ability to automatically learn. There have been many deep learning models based on 3D convolutional neural network (CNN) that have been proposed to solve this problem. In this paper, we propose a model which is called (2+1)D ResNet-18 to solve the falling detection task. The experimental results show that (2+1)D ResNet-18 gives 0.87% better accuracy on the FDD dataset and 1.13% on the URFD dataset than the recently proposed methods.