Emotion Recognition from Facial Expressions

Abstract/Technology Overview

Using advanced computer vision and artificial intelligence techniques, we provide a software package that analyzes facial expressions and reads emotions with high precision in real time from face images or video. This allows us to objectively analyse people’s reactions, behaviour, and experience.

Most other techniques for emotion recognition classify a face into one of 6 or 7 emotion categories. These traditional approaches are very limited and restrictive, because it is difficult to capture subtle expressions and distinguish some of the more complex emotions that people display in natural interactions. Our model offers an entirely new paradigm for continuous emotion/facial expression analysis in real-world conditions that overcomes these limitations.

Technology Features, Specifications and Advantages

Our software offers fine-grained estimations along 2 continuous emotion dimensions: valence (positive vs. negative emotions) and arousal (energetic vs. passive expressions). This approach is ideally suited for real-world applications because it can detect subtle emotions and differentiate between small variations in expressions, providing an objective and quantifiable measure for facial analytics. Our technology follows a new paradigm of facial expression analysis, which is proven to address the major issues of traditional approaches and exhibits the following advantages:

  • It is not limited to identifying specific predefined emotion categories, but can handle a wide range of different expressions in the continuous Arousal-Valence space.
  • It can estimate the intensity of the expression. As a result, it is also more successful in discriminating between low-intensity expressions.
  • The model has been trained with genuine facial expressions and not fake/posed ones. As such, estimations are more accurate in real world conditions.

Our algorithms achieve 90% and above accuracy for emotion recognition, having been trained with a database comprising hundreds of thousands of faces and expressions.

Potential Applications

  • Retail & Advertising: captures customer reactions, supports retail associates.
  • Human Resources: emotional analysis for pre-screening, assessment testing. More objective evaluation and ranking of candidates.
  • Education: mood meter for (virtual) classrooms; measures real-time learner responses and engagement; adapts and personalizes lessons; measures effectiveness of lecturer.
  • Transportation: monitors emotional states of drivers and passengers, integrated into driver assistance systems; increases safety, enhances passenger experience.
  • Healthcare: monitors emotional states and analyses mental well-being of patients or the elderly; integration with telemedicine platforms.
  • Entertainment: understands user emotions, adapts content for more immersive experiences.

Customer Benefit

Our software technology enables a fine-grained analysis of emotions and expressions, with instantaneous measurements for capturing spontaneous reactions, trends of emotional attributes over time, and aggregated statistics (e.g. comparative analytics between individuals, or crowd mood). High-level expression data analytics are customized for specific use cases. The software is light-weight and thus ideally suited for efficient implementations on an edge device or in the cloud.

Technology Owner

Songyou Peng


ADSC (Illinois at Singapore Pte Ltd)

Technology Category
  • Artificial Intelligence
  • Video/Image Analysis
Technology Status
  • Licensed but Available
Technology Readiness Level
  • TRL 7

emotion, computer vision, real time