EmotionSense — Real-Time Facial Emotion Recognition

MONITORING VIEWPORT

0.0 FPS

Camera Access Required

Click Start Monitoring below to initialize camera stream and begin real-time facial analysis.

SMOOTHING:

PRIVACY ASSURANCE: Video stream is processed in-memory in real-time. No image frames, facial recordings, or personal data are stored or transmitted to external servers.

PRIMARY CLASSIFICATION

STANDBY

0.0%

CONFIDENCE METRIC: 0.0000 INACTIVE

EMOTION PROBABILITY DISTRIBUTION

7-CLASS PROBABILITIES

ANGRY 0.0%

DISGUST 0.0%

FEAR 0.0%

HAPPY 0.0%

NEUTRAL 0.0%

SAD 0.0%

SURPRISE 0.0%

SESSION DIAGNOSTICS

INFERENCE TIME 0.0 ms

FRAME RATE 0.0 FPS

PROCESSED FRAMES 0

STREAM PROTOCOL WebSocket

SYSTEM SPECIFICATION & ARCHITECTURE

MODEL ACCURACY LOADING...

EMOTION CLASSES 7

INFERENCE RATE REAL-TIME (~10 FPS)

TECH STACK TENSORFLOW · OPENCV · FASTAPI

WHAT IT IS

EmotionSense is a healthcare-analytics-oriented real-time monitoring tool designed to detect human faces in video feeds and classify facial expressions across seven core emotional states: happy, neutral, sad, surprise, angry, fear, and disgust. Built with a restrained, clinical aesthetic, it provides low-latency confidence scoring and telemetry for physiological and observational research.

HOW IT WORKS

Webcam video frames are sampled at 5–10 FPS and processed through an OpenCV face detection pipeline to crop facial region-of-interest (ROI) tensors. The cropped 48×48 normalized grayscale image is evaluated by a 4-block Convolutional Neural Network (CNN) with Batch Normalization and Softmax output layers, producing real-time probability vectors smoothed via Exponential Moving Average (EMA).