Deep Learning Using for Fall Detection on the Rehabilitation Walking-Aid Robot
Biao Cao · 2019
Unexpected falls can result in severe injuries to disabled individuals who can barely maintain the balance of themselves. In this paper, a deep learning-based method for fall detection on the rehabilitation walking-aid robot was proposed. The goal of this method is to provide a real-time detector for the walking-aid robot so it can take preventive measures when the fall happened. The deep neural network with error back propagation to update the weight of each feature. The data set used in this paper was obtained upon the prototype of a walking-aid robot. The real-time motion data of the user's waist can be obtained by multiple sensors mounted on the walking-aid robot. Extract the motion features from the acquired data and tag it as normal or abnormal (potential to falling) feature, these two parts of data compose to a sample. Divide the whole data set into two parts to train and test the neural network. It turns out the accuracy of this trained neural network can be 95.8%. In the end, the effect of the parameter and hyper-parameters of the neural network was discussed.