METHOD AND APPARATUS FOR RECOGNIZING BEHAVIOR PATTERNS BY COMBINING MULTIPLE RECOGNITION MODELS

摘要

The present invention relates to a method for recognizing behavior patterns by combining multiple recognition models recognizing the behavior pattern of human in a behavior pattern recognizing apparatus, comprising the steps of: allowing recognition models to process image data for learning, including skeleton data for each behavior pattern input from a RGB-D camera sensor, into processing data for learning capable of recognizing the characteristics of each behavior pattern; learning the processing data for learning through a LSTM-based DNN recognition model and a SVM recognition model and storing learned values in a LSTM-based DNN recognition model database and a SVM recognition model database for each behavior pattern; allowing the recognition models to process the image data for recognition, including the skeleton data for each behavior pattern input from the RGB-D camera sensor, into input processing data for recognition capable of recognizing the characteristics of each behavior pattern; analyzing the input processing data for recognition through the LSTM-based DNN recognition model and the SVM recognition model to calculate a SVM probability value and a LSTM probability value; performing the recognition by performing a weighted sum operation of summing a value obtained by multiplying the LSTM probability value by a first weight and a value obtained by multiplying the SVM probability value by a second weight value and selecting a behavior class having the largest value as a behavior class pattern about an image for recognition. Thus, recognition rate of the human behavior pattern can be improved.;COPYRIGHT KIPO 2019