-
-
Real-time Violence Detection & Alert System with AI-Powered Surveillance and Live Geo-Mapping
🚀 Inspiration Violence in public spaces often goes undetected until it’s too late to respond effectively. With increasing CCTV coverage, there's an opportunity to leverage AI to detect real-time threats and dispatch alerts before lives are lost. The idea behind SVAS (Smart Violence Alert System) was to create a smart, automated solution that identifies violent acts from video feeds and instantly alerts authorities — combining AI, location tracking, and emergency response into one powerful system.
💡 What it does SVAS is an AI-powered system that detects real-life violence from camera feeds using deep learning. Once violence is detected:
An automatic alert is triggered.
The exact location is captured using IP-based geolocation.
A call is initiated to a predefined emergency contact (e.g., a control room).
The incident is displayed on a map in real time.
A log is created for trend analysis and response tracking.
It’s built to work with both live CCTV feeds and uploaded video clips.
🛠️ How we built it Model: A ResNet18 + LSTM hybrid deep learning model trained on the “Real-Life Violence Situations” dataset.
Backend: Python with Flask for video processing, detection logic, and service integration.
Frontend: Streamlit for the prototype interface, showing detection logs and map visuals.
Location Tracking: Geocoder for real-time IP-based location data.
Call Automation: Integrated Twilio-based service (or placeholder) to simulate emergency alerts.
Visualization: Folium and Streamlit-Folium for incident mapping.
The project follows a modular design with separated utils, models, and services directories for better structure and maintainability.
🧩 Challenges we ran into Getting real-time video analysis to run efficiently without frame loss or lag.
Balancing detection accuracy with performance (especially avoiding false positives).
Ensuring cross-platform compatibility for location tracking.
Creating a seamless integration between detection, location, call alerts, and frontend visualization.
🏆 Accomplishments that we're proud of Built and deployed a working prototype that detects violent actions with high accuracy.
Automated the entire alert pipeline — from detection to location tracking and call initiation.
Created a user-friendly frontend for non-technical stakeholders to monitor incidents.
Visualized incidents on a live map for better situational awareness.
📚 What we learned How to optimize deep learning models for video-based action recognition.
Integrating AI with real-world services like geolocation APIs and telephony systems.
The importance of latency, UX, and modularity in real-time safety applications.
Best practices in structuring ML projects for both scalability and readability.
🔮 What's next for SVAS (Smart Violence Alert System) Deploy to real CCTV feeds in collaboration with local authorities or institutions.
Integrate face recognition to identify repeat offenders.
Expand location tracking with GPS/mobile data for more accuracy.
Develop a mobile app version for on-the-go alerts and citizen reporting.
Add a dashboard for analytics, incident heatmaps, and predictive insights using time-based trends.
Log in or sign up for Devpost to join the conversation.