Real-Time People Counting Using MobileNet SSD

This project implements a real-time people counting system using a fixed webcam and a deep learning-based object detector. The system detects people entering or exiting a monitored area and tracks overall occupancy using a virtual line-crossing mechanism.

Features

• Live detection using MobileNet SSD (OpenCV DNN)
• Real-time occupancy tracking
• Horizontal or vertical line-crossing configuration
• Jitter-resistant centroid-based tracking
• Frame display with bounding boxes, IDs, and occupancy counters

Requirements

• Python 3.8+
• OpenCV (4.x)
• NumPy

Install requirements:

pip install opencv-python numpy

Running the System

Place the MobileNet SSD model files in a models/ folder:

• MobileNetSSD_deploy.prototxt.txt
• MobileNetSSD_deploy.caffemodel

Then run:

python main.py

Output

• Live video window showing:
  • Detected people with bounding boxes and centroids
  • Virtual line (red)
  • Entry/exit and current occupancy counts (top left corner)

Notes

• Default input resolution is 640x480.
• By default, the virtual line is vertical and centered.
• You can adjust line orientation and position in the script config.

Future Work

• Multi-person tracking improvements
• Multi-camera handoff support
• Re-identification for lost IDs

Author

Corey Floyd