Human Behaviour Recognition

A lightweight computer vision prototype for recognizing human actions and facial emotions from a live camera feed. The project explores rule-based behavior detection, landmark-driven analysis, and an interactive Streamlit dashboard without relying on a large training dataset.

Overview

The system is designed around two complementary signals:

• Body/pose analysis for actions such as sitting, standing, walking, and waving
• Facial analysis for simple emotion categories such as happy, angry, and surprised
• Interactive analytics for viewing detected behavior, emotion distribution, session activity, and performance information

The project is intended as a practical computer-vision prototype and learning project, with an emphasis on lightweight processing and interpretable detection rules.

Key Features

• Real-time camera input
• Pose and facial landmark processing
• Rule-based behavior detection
• Emotion classification using visual features
• Streamlit-based dashboard
• Session-level behavior and emotion analytics
• Activity logging and visualization

How It Works

The architecture below is intentionally centered so the main input and output stages line up with the combined processing branches:

                         Camera Input
                              │
                              ▼
                       Frame Processing
                              │
                 ┌────────────┴────────────┐
                 ▼                         ▼
           Pose Analysis              Face Analysis
                 │                         │
                 ▼                         ▼
          Behavior Rules             Emotion Rules
                 │                         │
                 └────────────┬────────────┘
                              ▼
                     Detection Results
                              │
                              ▼
                   Streamlit Dashboard
                              │
                              ▼
                    Session Analytics

Tech Stack

Technology	Purpose
Python	Application logic
OpenCV	Camera and image processing
Streamlit	Interactive application interface
Pandas	Data handling and analytics
Plotly	Visualization

Project Structure

.
├── app.py
├── cam_handler.py
├── pose_behavior.py
├── emotion_engine.py
├── dashboard_metrics.py
├── test_behavior.py
├── test_emotion.py
├── requirements.txt
└── README.md

My Contribution

Aarushi Agrawal — Analytics & Visualization

• Worked on the analytics and visualization layer
• Used Pandas for session-level data handling
• Used Plotly for presenting behavior and emotion distributions
• Contributed to testing, integration, and project documentation

What This Project Demonstrates

• Real-time computer-vision workflow design
• Image and camera-frame processing
• Landmark-based reasoning
• Rule-based classification
• Data collection and session analytics
• Building an interactive Python application

Project Status

This repository represents a lightweight prototype/academic project focused on demonstrating the computer-vision workflow and dashboard concept. The detection approach is intentionally rule-based rather than a production-grade trained recognition system.

Future Improvements

• Replace heuristic rules with trained models
• Add more robust validation and test coverage
• Improve detection under different lighting and camera conditions
• Add model-confidence calibration
• Containerize and deploy the application

Author

Aarushi Agrawal

• GitHub: CrapeBell
• LinkedIn: Aarushi Agrawal