AI-Based Public Safety Monitoring and Risk Detection System

This project is a full-stack AI-based public safety monitoring and risk detection system for crowd analysis, built for the Gemini 3 Hackathon.

The system combines computer vision with Gemini 3–powered reasoning to move beyond simple detection and enable context-aware public safety intelligence.

---

Project Structure

Public-Safety-Monitoring/
Core backend and frontend for crowd anomaly detection, contextual reasoning, and alerting

---

Key Idea

Traditional surveillance systems detect what is happening.
This system uses Gemini 3 to understand why it is happening and what action should be taken.

Computer vision models extract spatial and temporal signals from video, while Gemini 3 performs higher-level reasoning such as risk assessment, cause-and-effect analysis, and alert decision-making.

---

Features

Real-time and batch video analysis for crowd risk detection
Spatial-temporal crowd behavior analysis
Risk classification: NONE / LOW / MEDIUM / HIGH
Gemini 3–based contextual reasoning for risk severity and alert decisions
Automatic police alert creation for MEDIUM and HIGH risk events
User dashboard for video upload and risk timeline visualization
Police dashboard for alert monitoring and acknowledgment
Persistent alert storage
Modern React frontend and FastAPI backend

---

How Gemini 3 Is Used

Gemini 3 acts as the reasoning and decision-making layer of the system.

Workflow:

1. Video frames are processed using computer vision models to extract crowd density, motion patterns, and anomalies.
2. These signals are summarized into structured scene descriptions.
3. The summaries are sent to the Gemini 3 API.
4. Gemini 3 analyzes context across time to understand cause-and-effect relationships.
5. Gemini 3 classifies risk severity and decides whether an alert should be triggered.
6. Gemini 3 generates human-readable incident explanations for authorities.

This approach reduces false positives and transforms raw detections into actionable public safety intelligence.

---

Technologies Used

Backend
Python
FastAPI
OpenCV
TensorFlow
Keras
YOLOv8
Uvicorn
python-dotenv

Frontend
React
Vite
Tailwind CSS

AI and Reasoning
Google Gemini 3 API for contextual reasoning, risk assessment, and alert generation

Other
Node.js
REST APIs

---

Quick Start

Backend Setup

cd Public-Safety-Monitoring/backend
python -m venv .venv
.\.venv\Scripts\activate
pip install -r requirements.txt
uvicorn app.main:app --reload --host 0.0.0.0 --port 8000

Add your Gemini 3 API key in the .env file.

Frontend

cd Public-Safety-Monitoring/frontend
npm install
npm run dev

Folder Overview

Public-Safety-Monitoring/backend/ FastAPI backend, video analyzers, Gemini 3 integration, alert logic, and storage

Public-Safety-Monitoring/frontend/ React dashboards for users and police authorities

Public-Safety-Monitoring/Crowd_Anomaly_Detection/ Pretrained models and scripts for crowd analysis

Environment Variables

Required configuration files:

backend/.env.example frontend/.env

These include Gemini 3 API credentials and service configuration values.

Contributing

Pull requests are welcome. For major changes, please open an issue to discuss your proposal.