Activity recognition using inertial sensors and a 2-D convolutional neural network

Activity recognition has a growing interest in many fields like biomedical engineering, game development or for sports training. Sensors are attached to a human body to track body movement, physiological signals or environmental variables and these informations are interpreted by algorithms. The finding of characteristics of sensor data for the classification problem plays an important role and is commonly done by hand and with a lot of expertise by the researcher. In this paper, a 2-D Convolutional Neural Network (CNN) is used to extract features of images, which are fed in a Support Vector Machine (SVM) for classification. The key idea are spectrograms produced from 1-D signals of inertial sensors. To evaluate the proposed system, two datasets are used.