Daily Motion Recognition System by a Triaxial Accelerometer Usable in Different Positions


Kurban O. C., YILDIRIM T.

IEEE SENSORS JOURNAL, vol.19, no.17, pp.7543-7552, 2019 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 19 Issue: 17
  • Publication Date: 2019
  • Doi Number: 10.1109/jsen.2019.2915524
  • Journal Name: IEEE SENSORS JOURNAL
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus
  • Page Numbers: pp.7543-7552
  • Keywords: Motion recognition, accelerometer sensor, feature reduction, fall detection, human-computer interaction, RBF NEURAL-NETWORKS, CLASSIFICATION, FALL, IMPLEMENTATION, ALGORITHMS, MOVEMENT, POSTURE
  • Yıldız Technical University Affiliated: Yes

Abstract

Human-computer interaction has been a more popular research field in recent years with the escalating interest in wearable systems. In this context, evaluating the details of the physical movement of the human body has become a prominent subject for researchers. The systems, which are yielded from this subject, are utilized in such areas as health, public safety, access control, and smart home systems. This kind of system is expected to give fast and reliable results and can be used comfortably for the user. In this paper, a fast and accurate system design about recognition of daily human activities, including falls in the elderly, is proposed. In addition, another main objective of the study is to examine whether an accelerometer can be set in different positions during person's daily activities. Motion data have been acquired from the waist, wrist, and knee positions of 5 female and 15 male volunteers using a three-axis accelerometer. In this way, walking, sitting, standing up, jumping, and falling motions were collected from volunteers. The principal component analysis method was used as a feature reduction method to reduce the process time and get quick results. Furthermore, any feature extraction method was not used. The support vector machine, multi-layer perceptron, radial basis function, and Naive Hayes classifiers were employed to perform classification success. The proposed technique achieved a maximum of 100%, an average of 96.54% accuracy on the three defined positions. As a result, it has been determined that a single accelerometer can he set in different positions. The findings also suggest that this method can be used in home care and follow-up procedures and fall detection for elderly people.