| | --- |
| | title: Face Counter & Density Estimator |
| | emoji: 🎯 |
| | colorFrom: indigo |
| | colorTo: blue |
| | sdk: gradio |
| | sdk_version: 5.29.0 |
| | app_file: app.py |
| | pinned: false |
| | --- |
| | |
| | 🎯 Project Title: Face Counting and Crowd Density Estimation using MTCNN |
| | 🔍 Overview |
| | This project addresses the problem of estimating the number of faces in an image, with the added goal of classifying crowd density as Sparse, Medium, or Dense. Initially approached as a regression task, we shifted strategies after empirical evaluation showed that using ground-truth bounding box annotations from the WIDER FACE dataset provided more reliable face counts than a trained regressor. We ultimately adopted the MTCNN (Multi-task Cascaded Convolutional Networks) approach for real-time face detection due to its robustness, accuracy, and speed in varying crowd scenarios. |
| |
|
| | 🧠 Why MTCNN? |
| | MTCNN is a popular face detection framework because: |
| |
|
| | It combines face detection and facial landmark localization, making it suitable for fine-grained analysis. |
| |
|
| | It performs well across scales, poses, and lighting conditions. |
| |
|
| | It runs in real-time, enabling its use in live applications like webcam feeds. |
| |
|
| | It's pretrained and optimized, saving time and training resources. |
| |
|
| | 💡 How It Works |
| | A user uploads an image or uses their webcam to capture one. |
| |
|
| | The app runs MTCNN to detect faces and count bounding boxes. |
| |
|
| | Based on the count: |
| |
|
| | Sparse: 1–10 faces |
| |
|
| | Medium: 11–50 faces |
| |
|
| | Dense: 51+ faces |
| |
|
| | Results are overlaid directly on the image, offering a visual interpretation of the density. |
| |
|
| | 🛠️ Tech Stack |
| | Model: facenet-pytorch MTCNN |
| |
|
| | Dataset: WIDER FACE (for validation and benchmarking) |
| |
|
| | Framework: Python, Streamlit |
| |
|
| | Libraries: OpenCV, PIL, Torch, Matplotlib |
| |
|
| | Deployment: Streamlit (local or cloud-hosted) |
| |
|
| | 🌍 Real-Life Applications |
| | Surveillance & Public Safety |
| | Detect unusually dense crowds in public areas to trigger alerts. |
| |
|
| | Event Management |
| | Monitor real-time foot traffic and optimize crowd control in concerts, rallies, etc. |
| |
|
| | Retail Analytics |
| | Gauge customer distribution across zones in malls or stores. |
| |
|
| | Transportation Hubs |
| | Analyze crowd density in airports or stations to deploy personnel dynamically. |
| |
|
| | Smart Cities |
| | Integrated with CCTV systems, the app can be part of intelligent urban monitoring. |
| |
|
| | Pandemic Safety Enforcement |
| | Identify when crowd limits are exceeded to enforce health protocols. |
| |
|
| | ✅ Impact |
| | This project demonstrates how a lightweight, pretrained architecture like MTCNN can replace heavyweight regression-based models when data annotation is reliable. It bridges computer vision with social and safety applications, showing how academic tools can translate into real-world solutions. |
