<![CDATA[Smart motorcycle helmets are an emerging topic that can provide convenience to motorcyclists, such as providing information about the gas tank and tire pressure through the sound on the helmet. However, extracting important features of sequential data from accelerometer sensors becomes challenging when attempting to add a fall detection function to the helmet. This study proposes a gated recurrent unit (GRU) for fall detection using an accelerometer mounted on a smart motorcycle helmet. The first step is to get the x-axis, y-axis, and z-axis data from the accelerometer for the fallen human condition and the non-falling human condition. The data preparation involves the autocorrelation function (ACF), the partial autocorrelation function (PACF), normalization, standardization, random oversampling, and one hot encoder. The last is to train the GRU model. We use long short-term memory (LSTM) and convolutional neural network (CNN) as benchmarks. Accuracy, Loss, Precision, Recall, and F1−Score are the metrics we use to measure model performance. The test results show that GRU has Accuracy that is better than LSTM and CNN, which are 0.98, 0.97, and 0.96, respectively. Then other GRU performances in fall detection using the accelerometer sensor are 0.99, 0.97, and 0.98 for Precision, Recall, and F1−Score, respectively.]]>
Copyrights © 2022
