Activity recognition using consumer-grade EEG device

The human body, including brain, generates biological signals. The electroencephalogram, or EEG is the record of the brain electric activity, measured on the scalp. Nowadays acquisition of these signals is possible with affordable, consumer-grade devices. This paper proposes an approach to use an Ultracortex Mark IV for data acquisition and explore the variations in the EEG signals applying a convolutional neural network (CNN) when a person moves his or her fists or feet. Eight electrodes were placed over the head to record the cortical electrical activity. As a pre-processing step, detrending was used to remove slow drift from the time series data. Detrended data were used to train a fifteen-layer convolutional neural network. Achieved accuracy was 94% in classification of three different activities.