Every parent, teacher, and student wants safer schools. Cameras and other surveillance systems add an important layer of security but also raise serious privacy concerns. Researchers at the Meharry School of Applied Computational Sciences (SACS) have engineered a solution: a safety system that spots potential threats in real time—without streaming or storing students’ private videos on distant corporate servers.
Traditional campus surveillance relies on massive cloud networks. Video feeds are piped off-site to a centralized server, a process that risks student data breaches and slows down response times when every second counts.
The framework designed by Uttam Ghosh, Ph.D., professor of cybersecurity, and his team, shifts the intelligence directly to the school campus. By using edge computing, which runs AI models directly on local hardware installed inside the school rather than across the internet, the system processes multi-modal data locally and keeps sensitive video within school walls.
“This privacy-first approach demonstrates how data science can solve urgent public safety challenges while actively protecting civil liberties and student digital rights,” said Dr. Ghosh.
Key Technical Innovations:
- Zero Raw Video Leaves Campus: Video feeds from security cameras, authorized drones, environmental sensors, and audio devices are analyzed on-site. The AI automatically blurs faces and strips out personal identifiers right at the camera level.
- Federated Learning Across Campuses: Through federated learning—a technique where decentralized machines train an AI algorithm collaboratively without ever sharing their actual source data—multiple schools share only mathematical model “lessons” (weights), not student footage. One school’s system learns to detect a new hazard and immediately helps other schools recognize it too.
- Offline-First Resilience: If a campus loses internet connectivity or faces a network outage during an emergency, the safety AI does not shut down. Local edge servers continue weapon and anomaly detection uninterrupted, synchronizing updates once the connection is restored.
- Multi-Sensor Threat Recognition: Instead of relying only on basic video, the setup evaluates audio spikes, environmental changes, and dynamic overhead perspectives from authorized unmanned aerial vehicles (UAVs) to confirm risks before sounding false alarms.
Dr. Ghosh’s co-authors for the study are Eugene Levin, Ph.D., professor of spatial data science; Pushpita Chatterjee, Ph.D., assistant professor of computer science and data science; Debashis Das, Ph.D., postdoctoral fellow; David Lockett, grants proposal development and awards management specialist and Data Science Ph.D. student, and La Chiara Landrum, a data science Ph.D. student.
The study is published in 2026 International Conference on Intelligent Multimedia, Networking, and Security (IMNS) and is part of the project CAMPUS (Community Air Mobility for Public Utilization in Support of Healthcare) funded by a $742,286 grant.
Transform the Future of Responsible Technology
This privacy-preserving school safety is one of the many projects combining cybersecurity and AI at Meharry SACs. Learn how you can start using data science, AI and cybersecurity to solve real-world community challenges.
Do you want to learn to build ethical, life-saving computational systems to solve community challenges? Explore our graduate programs and research initiatives.
