skip to main content

Title: Deep Learning-Based Emotion Recognition from Real-Time Videos
We introduce a novel framework for emotional state detection from facial expression targeted to learning environments. Our framework is based on a convolutional deep neural network that classifies people’s emotions that are captured through a web-cam. For our classification outcome we adopt Russel’s model of core affect in which any particular emotion can be placed in one of four quadrants: pleasant-active, pleasant-inactive, unpleasant-active, and unpleasant-inactive. We gathered data from various datasets that were normalized and used to train the deep learning model. We use the fully-connected layers of the VGG_S network which was trained on human facial expressions that were manually labeled. We have tested our application by splitting the data into 80:20 and re-training the model. The overall test accuracy of all detected emotions was 66%. We have a working application that is capable of reporting the user emotional state at about five frames per second on a standard laptop computer with a web-cam. The emotional state detector will be integrated into an affective pedagogical agent system where it will serve as a feedback to an intelligent animated educational tutor.
Authors:
Award ID(s):
1821894
Publication Date:
NSF-PAR ID:
10276161
Journal Name:
Kurosu M. (eds) Human-Computer Interaction. Multimodal and Natural Interaction. HCII 2020. Lecture Notes in Computer Science
Volume:
12182
Sponsoring Org:
National Science Foundation
More Like this
  1. Affective studies provide essential insights to address emotion recognition and tracking. In traditional open-loop structures, a lack of knowledge about the internal emotional state makes the system incapable of adjusting stimuli parameters and automatically responding to changes in the brain. To address this issue, we propose to use facial electromyogram measurements as biomarkers to infer the internal hidden brain state as feedback to close the loop. In this research, we develop a systematic way to track and control emotional valence, which codes emotions as being pleasant or obstructive. Hence, we conduct a simulation study by modeling and tracking the subject'smore »emotional valence dynamics using state-space approaches. We employ Bayesian filtering to estimate the person-specific model parameters along with the hidden valence state, using continuous and binary features extracted from experimental electromyogram measurements. Moreover, we utilize a mixed-filter estimator to infer the secluded brain state in a real-time simulation environment. We close the loop with a fuzzy logic controller in two categories of regulation: inhibition and excitation. By designing a control action, we aim to automatically reflect any required adjustments within the simulation and reach the desired emotional state levels. Final results demonstrate that, by making use of physiological data, the proposed controller could effectively regulate the estimated valence state. Ultimately, we envision future outcomes of this research to support alternative forms of self-therapy by using wearable machine interface architectures capable of mitigating periods of pervasive emotions and maintaining daily well-being and welfare.« less
  2. The overall goal of our research is to develop a system of intelligent multimodal affective pedagogical agents that are effective for different types of learners (Adamo et al., 2021). While most of the research on pedagogical agents tends to focus on the cognitive aspects of online learning and instruction, this project explores the less-studied role of affective (or emotional) factors. We aim to design believable animated agents that can convey realistic, natural emotions through speech, facial expressions, and body gestures and that can react to the students’ detected emotional states with emotional intelligence. Within the context of this goal, themore »specific objective of the work reported in the paper was to examine the extent to which the agents’ facial micro-expressions affect students’ perception of the agents’ emotions and their naturalness. Micro-expressions are very brief facial expressions that occur when a person either deliberately or unconsciously conceals an emotion being felt (Ekman &Friesen, 1969). Our assumption is that if the animated agents display facial micro expressions in addition to macro expressions, they will convey higher expressive richness and naturalness to the viewer, as “the agents can possess two emotional streams, one based on interaction with the viewer and the other based on their own internal state, or situation” (Queiroz et al. 2014, p.2).The work reported in the paper involved two studies with human subjects. The objectives of the first study were to examine whether people can recognize micro-expressions (in isolation) in animated agents, and whether there are differences in recognition based on the agent’s visual style (e.g., stylized versus realistic). The objectives of the second study were to investigate whether people can recognize the animated agents’ micro-expressions when integrated with macro-expressions, the extent to which the presence of micro + macro-expressions affect the perceived expressivity and naturalness of the animated agents, the extent to which exaggerating the micro expressions, e.g. increasing the amplitude of the animated facial displacements affects emotion recognition and perceived agent naturalness and emotional expressivity, and whether there are differences based on the agent’s design characteristics. In the first study, 15 participants watched eight micro-expression animations representing four different emotions (happy, sad, fear, surprised). Four animations featured a stylized agent and four a realistic agent. For each animation, subjects were asked to identify the agent’s emotion conveyed by the micro-expression. In the second study, 234 participants watched three sets of eight animation clips (24 clips in total, 12 clips per agent). Four animations for each agent featured the character performing macro-expressions only, four animations for each agent featured the character performing macro- + micro-expressions without exaggeration, and four animations for each agent featured the agent performing macro + micro-expressions with exaggeration. Participants were asked to recognize the true emotion of the agent and rate the emotional expressivity ad naturalness of the agent in each clip using a 5-point Likert scale. We have collected all the data and completed the statistical analysis. Findings and discussion, implications for research and practice, and suggestions for future work will be reported in the full paper. ReferencesAdamo N., Benes, B., Mayer, R., Lei, X., Meyer, Z., &Lawson, A. (2021). Multimodal Affective Pedagogical Agents for Different Types of Learners. In: Russo D., Ahram T., Karwowski W., Di Bucchianico G., Taiar R. (eds) Intelligent Human Systems Integration 2021. IHSI 2021. Advances in Intelligent Systems and Computing, 1322. Springer, Cham. https://doi.org/10.1007/978-3-030-68017-6_33Ekman, P., &Friesen, W. V. (1969, February). Nonverbal leakage and clues to deception. Psychiatry, 32(1), 88–106. https://doi.org/10.1080/00332747.1969.11023575 Queiroz, R. B., Musse, S. R., &Badler, N. I. (2014). Investigating Macroexpressions and Microexpressions in Computer Graphics Animated Faces. Presence, 23(2), 191-208. http://dx.doi.org/10.1162/

    « less
  3. he goal of this research is to identify key affective body gestures that can clearly convey four emotions, namely happy, content, bored, and frustrated, in animated characters that lack facial features. Two studies were conducted, a first to identify affective body gestures from a series of videos, and a second to validate the gestures as representative of the four emotions. Videos were created using motion capture data of four actors portraying the four targeted emotions and mapping the data to two 3D character models, one male and one female. In the first study the researchers identified body gestures that aremore »commonly produced by individuals when they experience each of the four emotions. In the second study the researchers tested four sets of identified body gestures, one set for each emotion. The animated gestures were mapped to the 3D character models and 91 participants were asked to identify the emotional state conveyed by the characters through the body gestures. The study identified six gestures that were shown to have an acceptable recognition rate of at least 80% for three of the four emotions tested. Contentment was the only emotion which was not conveyed clearly by the identified body gestures. The gender of the character had a significant effect on recognition rates across all emotions.« less
  4. Medical image segmentation is one of the most challenging tasks in medical image analysis and widely developed for many clinical applications. While deep learning-based approaches have achieved impressive performance in semantic segmentation, they are limited to pixel-wise settings with imbalanced-class data problems and weak boundary object segmentation in medical images. In this paper, we tackle those limitations by developing a new two-branch deep network architecture which takes both higher level features and lower level features into account. The first branch extracts higher level feature as region information by a common encoder-decoder network structure such as Unet and FCN, whereas themore »second branch focuses on lower level features as support information around the boundary and processes in parallel to the first branch. Our key contribution is the second branch named Narrow Band Active Contour (NB-AC) attention model which treats the object contour as a hyperplane and all data inside a narrow band as support information that influences the position and orientation of the hyperplane. Our proposed NB-AC attention model incorporates the contour length with the region energy involving a fixed-width band around the curve or surface. The proposed network loss contains two fitting terms: (i) a high level feature (i.e., region) fitting term from the first branch; (ii) a lower level feature (i.e., contour) fitting term from the second branch including the (ii1) length of the object contour and (ii2) regional energy functional formed by the homogeneity criterion of both the inner band and outer band neighboring the evolving curve or surface. The proposed NB-AC loss can be incorporated into both 2D and 3D deep network architectures. The proposed network has been evaluated on different challenging medical image datasets, including DRIVE, iSeg17, MRBrainS18 and Brats18. The experimental results have shown that the proposed NB-AC loss outperforms other mainstream loss functions: Cross Entropy, Dice, Focal on two common segmentation frameworks Unet and FCN. Our 3D network which is built upon the proposed NB-AC loss and 3DUnet framework achieved state-of-the-art results on multiple volumetric datasets.« less
  5. Objective: The rapid growth of online health social websites has captured a vast amount of healthcare information and made the information easy to access for health consumers. E-patients often use these social websites for informational and emotional support. However, health consumers could be easily overwhelmed by the overloaded information. Healthcare information searching can be very difficult for consumers, not to mention most of them are not skilled information searcher. In this work, we investigate the approaches for measuring user similarity in online health social websites. By recommending similar users to consumers, we can help them to seek informational and emotionalmore »support in a more efficient way. Methods: We propose to represent the healthcare social media data as a heterogeneous healthcare information network and introduce the local and global structural approaches for measuring user similarity in a heterogeneous network. We compare the proposed structural approaches with the content-based approach. Results: Experiments were conducted on a data set collected from a popular online health social website,and the results showed that content-based approach performed better for inactive users, while structural approaches performed better for active users. Moreover, global structural approach outperformed local structural approach for all user groups. In addition, we conducted experiments on local and global structural approaches using different weight schemas for the edges in the network. Leverage performed the best for both local and global approaches. Finally, we integrated different approaches and demonstrated that hybrid method yielded better performance than the individual approach. Conclusion: The results indicate that content-based methods can effectively capture the similarity of inactive users who usually have focused interests, while structural methods can achieve better performance when rich structural information is available. Local structural approach only considers direct connections between nodes in the network, while global structural approach takes the indirect connections into account. Therefore, the global similarity approach can deal with sparse networks and capture the implicit similarity between two users. Different approaches may capture different aspects of the similarity relationship between two users. When we combine different methods together, we could achieve a better performance than using each individual method.« less