How-to · Displai
Use emotion recognition (Sightcorp)
Interpret the seven emotion scores DeepSight Toolkit reports for each detected face, and get more accurate emotion results.
When to use this
Section titled “When to use this”Use emotion recognition to analyze the emotional state of faces detected by the DeepSight Toolkit. This feature detects seven emotions and helps you understand customer engagement and sentiment in your physical spaces.
- Ensure you are running DeepSight Toolkit 7.0 or later.
- Enable the Sightcorp emotion recognition feature in your Displai settings.
- View emotion scores for each detected face. The system reports seven emotions: neutral, anger, disgust, fear, happiness, sadness, and surprise.
- Each emotion receives a score between 0 and 1, and all seven scores always add up to 1.0 at any point in time.
Verify
Section titled “Verify”Check that emotion scores appear in your analytics dashboard. Verify that all seven emotion scores for a given face sum to 1.0.
Interpreting emotion scores
Section titled “Interpreting emotion scores”Report the dominant emotion
Section titled “Report the dominant emotion”Report the highest average emotion score as the dominant emotion only if that average exceeds 0.5. For example:
- neutral = 0.0 (avg)
- anger = 0.1 (avg)
- disgust = 0.1 (avg)
- fear = 0.1 (avg)
- happiness = 0.5 (avg)
- sadness = 0.1 (avg)
- surprise = 0.1 (avg)
In this case, report the person as “happy” because happiness = 0.5.
Handle ambiguous results
Section titled “Handle ambiguous results”If all scores are low or too close to each other, categorize the person as “neutral”. For example:
- neutral = 0.2 (avg)
- anger = 0.1 (avg)
- disgust = 0.1 (avg)
- fear = 0.1 (avg)
- happiness = 0.2 (avg)
- sadness = 0.2 (avg)
- surprise = 0.1 (avg)
Report the dominant emotion as “none (neutral)” because no single emotion clearly dominates.
Accuracy considerations
Section titled “Accuracy considerations”Head pose and lighting affect emotion detection accuracy. For best results:
- Only use emotion estimations for faces detected within ±15 degrees of pitch and yaw (frontal faces).
- Exclude faces with extreme angles or poor lighting from emotion analysis.
If it doesn’t work
Section titled “If it doesn’t work”- Emotion scores seem inaccurate: Verify that faces are within ±15 degrees pitch and yaw. Check lighting conditions in your space.
- No emotion data appears: Confirm DeepSight Toolkit version is 7.0 or later and that Sightcorp integration is enabled.
- Privacy or compliance concerns: Contact support to discuss emotion recognition data handling and retention policies.
Related articles
Section titled “Related articles”- Does face detection work with beards, sunglasses, or eyeglasses? See how beards, eyeglasses, and sunglasses affect Sightcorp face detection, and why sunglasses can reduce emotion accuracy.
- Camera lighting requirements for Displai audience analytics Set up even, frontal lighting and avoid backlighting so your camera can detect faces accurately for audience analytics.
- Sightcorp privacy compliance Understand how Sightcorp audience analytics collects anonymous data without storing images or personally identifiable information.
Still having trouble?
Text the 24/7 line or email support.