Abstract:
PURPOSE: A device and method for recognizing a user activity in real time are provided to accurately recognize the user activity in real time regardless of how a user is holding a mobile terminal, thereby applying the same to various fields such as device control for the mobile terminal, user activity analysis, life logging, and user context recognition. CONSTITUTION: A device for recognizing a user activity in real time includes a learning data collecting part (150), an activity feature extraction model database (DB) (142a), an activity pattern classification model DB (143a), and a user activity recognizing part (140). The learning data collecting part collects a frequency domain signal by the user activity, and generates learning data. The activity feature extraction model DB stores an activity feature extraction model learned by a learning server based on the learning data. The activity pattern classification model DB stores an activity pattern classification model using the activity feature extraction model. The user activity recognizing part extracts a user activity feature and classifies a user activity pattern on the basis of the activity feature extraction model DB and the activity pattern classification model DB. [Reference numerals] (110) First sensor; (120) Second sensor; (130) Signal normalization unit; (141) Fast Fourier alteration unit; (142) Behavior property extraction unit; (142a) Behavior property extraction model DB; (143) Behavior pattern classification unit; (143a) Behavior pattern classification model DB; (150) Learning data collecting unit; (200) Learning server
Abstract:
Disclosed are a multi sensor and level information based user behavior recognition apparatus, an ONMF-based basis matrix generation method, and an OSSNMF-based basis matrix generation method. The user behavior recognition apparatus according to an aspect comprises: a feature vector extraction part for multiplying a sensor data matrix for sensor data of a frequency domain acquired from two or more sensors by a transposed matrix of an orthogonalized basis matrix to extract an ONMF-based feature vector; and a multi class classification part for classifying the extracted ONMF-based feature vector as one among two or more user behavior types.