Inspiration

Crowd safety is a major concern at public gatherings, railway stations, stadiums, festivals, and other crowded places. Traditional CCTV systems record video, but identifying dangerous crowd levels often requires continuous human monitoring. We wanted to make surveillance smarter by using AI to detect people, monitor crowd levels, and generate timely warnings.

What It Does

CrowdShield AI is an AI-powered crowd safety monitoring prototype that uses computer vision to detect and count people in video footage. It classifies crowd levels into three categories:

SAFE: 0–5 people
MODERATE: 6–9 people
HIGH: 10 or more people

When the detected crowd reaches the HIGH threshold, the system generates warnings. It also produces an annotated video, crowd statistics, a text report, and timeline data. A Streamlit dashboard presents the results through a monitoring interface with crowd analytics and alerts.

How We Built It

We developed CrowdShield AI using Python, YOLO11n, OpenCV, and Streamlit. YOLO11n performs person detection, OpenCV processes and annotates video frames, and Python calculates crowd statistics and manages alerts. We implemented a smart alert mechanism to reduce repeated warnings during a continuous high-crowd event. The dashboard displays the analysis results, while generated reports and JSON timeline data support further analysis.

Challenges We Faced

One of our main challenges was processing video frame by frame while maintaining reliable output and useful performance. Initially, the system generated repeated warnings for consecutive frames during the same high-crowd situation. We improved the alert logic to identify distinct HIGH crowd events instead of treating every frame as a separate event. We also worked on integrating the detection pipeline with the Streamlit dashboard and handling video inputs and generated output files.

What We Learned

Building CrowdShield AI helped us gain practical experience with computer vision, object detection, video processing, event-based alert logic, data reporting, and dashboard development. We learned how to connect an AI model to a usable application and turn raw detection results into information that can support human decision-making.

What's Next

Our next steps include browser-based live camera monitoring, configurable crowd thresholds, crowd-density heatmaps, multi-camera support, and analysis of crowd movement trends. We aim to develop CrowdShield into a more capable crowd-monitoring platform that helps event organizers and safety teams identify potential overcrowding earlier.