Service Robot’s Affective Computing Based On Facial Expression Cognition

Robot’s emotional recognition and humanoid emotions generation have received widespread attention in the research field of computer science and psychology. Meanwhile, they have been highly attached importance and got excellent support, then gradually formed an interdiscipline cognitive affective computing. This dissertation was focused on service robot’s cognitive emotional regulation in human-robot interaction. First, according to information-processing theories and biometrics the analysis model of visual attention was set up. As the cognitive basis in follow-up emotional regulation study, expression-emotion mapping model that could map an external expression stimulus into arousal-valence-stance3D emotional space was built with participant’s facial expression and micro-expression feature extraction. Second, based on situation selection, situation amendment and attention distribution in Gross cognitive emotional strategies, an emotional regulation and expression algorithm toward the psychology was proposed, and it was applied to humanoid service robot for interaction effects. Third, aimed to cognitive reappraisal and expressive suppression in Gross strategies, according to active field theory the emotional space that had dynamic performance was set up with the motivation psychology as the model sample. Emotional transfer model with HMM and Glauber power system was used for imitating human cognition and affection. Finally, put the above theories and algorithms to service robot and autism assisted therapy for experimental analysis and validation.The main contents and innovation points are as follows:(1) First, based on information-processing theories, visual attention’s asymmetric translocation in different salient regions was described by Markov chain-entropy. Second, according to facial action units in FACS, this paper proposed an expression emotion cognition method based on9facial feature region including13action units, and Gabor coefficients from5-scale and8-direction Gabor wavelet transform were trained by PCA for dimensionality reduction. Then atypical facial expression was mapped into various basic emotional states by expression potential for expressive emotion cognition. Third, the feature region in micro-expression key frame acquired from3D gradient projection was factored by Gabor filter, and the features decreased the dimension by LGBP for keep the characteristic information as far as possible. Then micro-expression feature was classified by weighted gradient NNA to capture participant’s underlying emotion. Finally, the above visual attention, facial expression and micro-expression emotion analytical method were the cognitive basis of affective computing.(2) First, interactive relationship and emotional suppression parameters were abstracted out from situation selection, situation amendment and attention distribution in Gross strategies for influencing the emotional state transfer probability. Second, the interactive relationship influence factor and emotional suppression influence factor were optimized by GA in order to improve the control accuracy. Then an emotional regulation and expression model with finite-state machine was formed and applied for humanoid service robot’s emotional expression combined with some external environment factors. Finally, the above model was used for human robot interaction. Compared with traditional model,14DOF humanoid service robot with cognitive emotional regulation model is more intelligent and reality, so it can give full play to its emotional diversification in the interaction.(3) First, service robot’s continuous and controlled emotional regulation was built in active field state space based on field force and conflict theory in human-robot interaction. Second, according to Fechner-Weber law emotional spontaneous and guidable cognitive reappraisal in Gross strategies was in quantitative analysis. Then emotional attenuation model associated with valence was set up by emotion intensity third low, and energy space in active field was proposed for simulating the interaction between external stimulus and robot’s emotion. Finally, with this research, the stimulating emotional transfer algorithm based on HMM and the spontaneous emotional transfer algorithm based on Glauber power system were came up to realize real-time dynamic control on13DOF service robot for uncertainty emotional regulation in interaction with atypical facial expressions.(4) On the research foundation of visual attention, expression/micro-expression emotion mapping and cognitive emotional regulation, a social assisted therapy system centered on emotional robot for children with autism was designed and it had been applied for clinical test in order to make up for relieving autism children’s cognition and emotion deficits and completing children-robot emotional interaction therapy. The visual attention model was devoted to joint attention subsystem for autism training. The expression/micro-expression emotion mapping model was used to expression learning and imitation training for real-time facial expression recognition, moreover it could trace children’s emotion in whole training. And robot could show corresponding behaviors based on cognitive emotional regulation. In scene cognition test training robot used emotional regulation model based on MFCCs for realizing emotional behavior. The model and algorithm proposed in this paper lined through the interaction between autism child and emotional robot. In addition, the mental and behavior interaction was implemented in the clinical test. SRS scale results show that cognitive emotional training with six modes can effectively improve children’s sociability more than the traditional rehabilitation.