Skip to main content
. 2020 Jun 5;20(11):3207. doi: 10.3390/s20113207

Figure 8.

Figure 8

CNN-based AI algorithm used to predict the risk of fall. The input layer receives a 20 s frame from the 6 inputs of the DYSKIMOT sensor. The convolutional layers (Conv2D layer) extract meaningful features from its input and present it to the next stage. Pooling layers (MaxPool2D Layer) reduce the dimension of the feature space by retaining the meaningful inputs of its layer only. After several convolutional stages, the remaining features are injected in a three layers neural network (dense layer) in order to classify the participant’s frames (output layer). The risk of fall is the mean (ranging from 0 to 1) of prediction on all frames of the participant’s sequence. A mean greater than 0.5 denotes a participant with risk of fall.