FEATURE REPRESENTATION EXTRACTING METHOD FOR RECONIZING HUMAN ACTIVITY AND ACTIVITY RECOGNITION APPRATUS THEREOF

摘要

Disclosed are a feature information extraction method for human behavior recognition and a behavior recognition device thereof. According to an embodiment of the present invention, the feature information extraction method comprises: a step of receiving sensing data from an acceleration sensor and a direction sensor to generate a first feature set array; a step of sequentially moving a first window in the first feature set array to generate a first window average array based on the average value in the first window; a step of combining the first feature set array with the first window average array to produce a second feature set array; a step of sequentially moving a second window in a third feature set array generated by rearranging the second feature set array to generate a second window average array based on the average value in the second window; a step of combining the third feature set array with the second window average array to produce a fourth feature set array; and a step of outputting the fourth feature set array to a learning module in units of rows. Thus, human behavior recognition accuracy can be improved.