Inspiration
Every week, an average of two school shootings occur in the U.S., leaving students and staff in a constant state of fear. The lack of real-time threat detection means that potential dangers often go unnoticed until it is too late. Inspired by the urgent need for proactive security in schools and public spaces, we created ShieldVision—an AI-powered surveillance system that detects threats before they escalate. Our goal is simple: save lives by enabling faster responses to active threats.
What it does
ShieldVision is a real-time AI-powered security system that analyzes live CCTV footage to detect weapons and suspicious activities instantly. It ensures faster response times by:
- Detecting and classifying threats like guns, knives, and suspicious behavior using YOLOv8 and OpenCV.
- Alerting security teams and first responders with precise location details and threat classification.
- Sending emergency notifications to students, staff, and law enforcement for immediate action.
- Providing a real-time dashboard for security personnel to track and validate potential threats.
How we built it
We designed ShieldVision with a robust tech stack to ensure speed, accuracy, and scalability:
- Frontend – Next.js, TailwindCSS, ShadCN for an intuitive security dashboard.
- Backend – FastAPI to handle real-time alerts and communication.
- AI Threat Detection – YOLOv8 (PyTorch + OpenCV) to detect and classify weapons.
- Database – ChromaDB for metadata storage and retrieval.
- Real-time Processing – Multithreaded processing to analyze multiple CCTV feeds simultaneously.
To optimize performance, we use YOLO embeddings to filter and process only high-risk frames, reducing unnecessary computation while ensuring accuracy.
Challenges we ran into
- High computational demand – Processing multiple video streams in real time required multithreading and optimization to balance speed and accuracy.
- Mapping security zones – We had to manually map locations for precise threat localization.
- Limited resources for testing – Finding publicly available realistic datasets for weapon detection was a challenge.
- Minimizing false alarms – We had to ensure accurate detection to prevent unnecessary panic while maintaining sensitivity to real threats.
Accomplishments that we're proud of
- Successfully built a fully functional prototype in 36 hours with live video threat detection.
- Implemented real-time alerts that provide accurate threat type, location, and recommended action.
- Developed a low-cost yet scalable AI model that runs efficiently on local GPUs while leveraging the power of large models for validation.
- Integrated a user-friendly security dashboard to streamline threat monitoring for security personnel.
What we learned
- Many students and staff feel unsafe due to the rising number of school shootings.
- Existing surveillance systems lack proactive AI-powered detection, leading to delayed responses.
- Using pre-screened YOLO embeddings allows for fast, cost-effective threat detection.
- AI-powered security must balance accuracy and efficiency to be truly effective in high-risk environments.
What's next for ShieldVision
- Deployment in real-world settings – We aim to partner with schools, malls, and public spaces to integrate ShieldVision into existing security infrastructures.
- Improved AI accuracy – Refining our models with more diverse real-world datasets to improve detection reliability.
- Live integration with 911 and emergency response teams for automated crisis management.
- Expanding detection capabilities – Beyond weapons, we aim to identify aggressive behavior, unauthorized access, and crowd anomalies to prevent potential threats before they escalate.
ShieldVision is not just an AI model—it is a mission to create safer spaces for everyone.
Log in or sign up for Devpost to join the conversation.