Negative emotions sensitive humanoid robot with attention-enhanced facial expression recognition network
Document Type
Article
Publication Date
1-1-2022
Abstract
Lonely older adults and persons restricted in movements are apt to cause negative emotions, which is harmful to their mental health. A humanoid robot with audiovisual interactions is presented, which can correspondingly output positive facial expressions to relieve human's negative facial expressions. The negative emotions are identified through an attention-enhanced facial expression recognition (FER) network. The network is firstly trained on MMEW macro-and micro-expression databases to discover expression-related features. Then, macroexpression recognition tasks are performed by fine-tuning the trained models on several benchmarking FER databases, including CK+ and Oulu-CASIA. A transformer network is introduced to process the sequential features engineered by the FER network and output a final stable control order. This order is used to control the robot's facial motor units to generate different expressions, e.g., a smile expression. Evaluations on benchmarking databases verify that the proposed method can precisely recognize facial expressions. The joint modulation with the humanoid robot proves that the robot can respond effectively to the user's negative emotions.
Keywords
Humanoid robot, Facial expression recognition, Attention mechanism, Transfer learning, Negative emotions
Divisions
ai
Funders
National key R&D program of China [Grant No: 2018AAA0100800],Key Research and Development Program of Jiangsu [Grant No: BK20192004B & BE2018004-04],Guangdong Forestry Science and Technology Innovation Project [Grant No: 2020KJCX005],International Cooperation and Exchanges of Changzhou [Grant No: CZ20200035]
Publication Title
Intelligent Automation and Soft Computing
Volume
34
Issue
1
Publisher
Tech Science Press
Publisher Location
871 CORONADO CENTER DR, SUTE 200, HENDERSON, NV 89052 USA