Automated Categorization of Behavioral Quality Through Deep Neural Networks

The possibility to automatically judge the quality of the motor behavior of human subjects could permit to monitor elder people in their natural home environment. More specifically, it might allow to identify critical situations requiring medical intervention in real time and extract data supporting medical diagnosis. In this work, we present an AI system automatically evaluating the quality level of the walking behavior of human subjects. The system is composed of a depth camera, a skeleton tracking software, and a neural network. The network is trained by using a training set containing the behavioral data of behaviors of different quality displayed by different individuals. The quality of the behaviors included in the training set is manually evaluated by us. The preliminary results collected by using a small training set indicate that the system is able to judge the behavioral quality of new subjects, not included in the training set, reasonably well. In future works the system will be improved by collecting and using a larger training set and considering also behaviors of other classes.