딥러닝을 이용한 연속 시간 감정 상태 추론 시스템 설계

초록

Facial expressions are one of the efficient methods of expressing emotions. Most of the existing facial expression analysis techniques recognize emotional state only in still images. This study proposes a deep learning system for estimating emotional states in continuous time. The proposed emotional state estimation system in continuous time consists of three processes. Feature extraction using Convolutional Neural Network (CNN) model, Activation Map (AM)-based facial region detection and region of interest (RoI) Pooling, and sequential estimation using Recurrent Neural Network (RNN) model. In this paper, we performed an experiment to estimate the emotional state from in short video clips of facial expressions. Experimental results show that the proposed system can perform facial region detection faster than conventional CNN - based object detection models and emotional state estimation in continuous time.

키워드