We use MetaWearC Streaming Sensor to collect the real time data. It contains the data collected from the sensor of accelerometer and the sensor of gyroscope. In order to get more precise data, we deal with the data with either up-sampling or down-sampling according to the raw data we collect.
To avoid the length limitation and disturbance of connection of wire, the sensor board and cellphone are connected by Bluetooth Low Energy wireless technology. Compared with wifi, BLE consumes much less energy and has enough bandwidth for transmission the real time data. The MetaWearC board uses nRF51822 SOC from Nordic which contains BLE. And due to the low energy consumption of BLE, the board can be powered by CR2032 Coin cell battery.
Deep learning is a class of machine learning algorithms that use multiple non-linear layers. Each layer extract information from previous layer and finally reach the task such as image classification and others. Our model use traditional deep neural network which contain 4 hidden layer and one softmax output layer. Except the output layer, other layers use Relu as activation function. To avoid the problem of overfitting, we add dropout of 0.1 in the last two layers.