People Counting in Real-Time Using OpenCV

Introduction

This project provides a real-time people counting system using a live video stream or IP camera with OpenCV. The primary objective is to enable real-time footfall analysis for stores, buildings, shopping malls, etc., and to trigger alerts when the number of people exceeds a certain limit. This system can be particularly useful for ensuring compliance with occupancy limits, such as those enforced during the COVID-19 pandemic.

The system uses a Single Shot Detector (SSD) with a MobileNet architecture for efficient and fast object detection. The Centroid Tracker is employed to track detected individuals across frames, ensuring accurate counting.

Features

• Real-time people counting using live video streams or IP cameras.
• Alert system to notify staff when occupancy limits are exceeded.
• Optimized for performance with threading.
• Scalable for business use in various environments.
• Footfall analysis and reporting.

Implementation Details

SSD with MobileNet

• SSD (Single Shot MultiBox Detector): A fast and efficient object detection model that generates region proposals and detects objects in a single pass.
• MobileNet: A lightweight deep neural network designed for mobile and embedded vision applications, making it suitable for real-time processing on resource-constrained devices.

Centroid Tracker

• Centroid Calculation: Computes the centroid of the bounding boxes detected by the SSD.
• Object Tracking: Assigns a unique ID to each detected individual and tracks them across frames by updating the centroid positions.

Threading

• Threading Optimization: Utilizes threading to handle video frame processing and object detection in parallel, improving real-time performance.

Results and Performance

The system demonstrates efficient real-time people counting with high accuracy. The combination of SSD with MobileNet provides a good balance between speed and accuracy, while the Centroid Tracker ensures reliable tracking of individuals across frames.

Future Work

• Implement additional optimizations for performance improvements.
• Explore alternative object detection models for enhanced accuracy.
• Develop a web dashboard for real-time monitoring and reporting.
• Integrate with IoT devices for automated alerting and response systems.

Acknowledgements

• The project builds upon the work of Adrian Rosebrock and his tutorial on OpenCV People Counter.
• Special thanks to the open-source community for providing pre-trained models and valuable resources.