A Novel Method for Automatic Detection and Classification of Movement Patterns in Short Duration Playing Activities

• Published in: IEEE access 2018, Vol.6, p.53409-53425
• Main Authors: Rivera, Diego, Cruz-Piris, Luis, Fernandez, Susel, Alarcos, Bernardo, García, Antonio, Velasco, Juan R.
• Format: Article
• Language: English
• Subjects: Acceleration; Activity recognition; Classification; classification algorithms; Feature extraction; Hidden Markov models; Internet of Things; Legged locomotion; Machine learning; Motion perception; pattern recognition; Performance evaluation; Reference signals; sensor systems and applications; Sequences; Support vector machines; Toy manufacturing industry
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2018.2871732

Autonomous devices able to evaluate diverse situations without external help have become especially relevant in recent years because they can be used as an important source of relevant information about the activities performed by people (daily habits, sports performance, and health-related activities). Specifically, the use of this kind of device in childhood games might help in the early detection of developmental problems in children. In this paper, we propose a method for the detection and classification of movements performed with an object, based on an acceleration signal. This method can automatically generate patterns associated with a given movement using a set of reference signals, analyze sequences of acceleration trends, and classify the sequences according to the previously established patterns. This method has been implemented, and a series of experiments has been carried out using the data from a sensor-embedded toy. For the validation of the obtained results, we have, in parallel, developed two other classification systems based on popular techniques, i.e., a similarity search based on Euclidean distances and machine-learning techniques, specifically a support vector machine model. When comparing the results of each method, we show that our proposed method achieves a higher number of successes and higher accuracy in the detection and classification of isolated movement signals as well as in sequences of movements.