Back to Projects
2025

AI-Based Drone Surveillance System

Real-time computer vision platform with YOLOv8 detection and DeepSORT tracking

AI-Based Drone Surveillance System - Screenshot 1
AI-Based Drone Surveillance System - Screenshot 2
AI-Based Drone Surveillance System - Screenshot 3

Overview

Production-grade surveillance system combining YOLOv8 object detection (50+ mAP50-95), DeepSORT multi-object tracking (95%+ ID persistence), and full-stack analytics platform. Features real-time video processing at 30+ FPS, polygonal geofencing, severity-based alerting, and comprehensive audit logging. Built with Flask REST API (20+ endpoints), React 18 frontend, and JWT role-based access control. Deployed for critical infrastructure monitoring, border security, and event management.

The Problem

Traditional surveillance systems cannot process live drone footage with actionable object detection, tracking, and spatial analysis in real-time. Security operations require immediate threat identification, persistent object tracking across frames, automated geofence breach detection, and comprehensive audit trails. Manual monitoring is labor-intensive, error-prone, and cannot scale to multi-camera deployments. Existing solutions lack customizable alerting, historical analytics, and secure multi-user access controls necessary for enterprise security operations.

The Solution

Built a production-grade computer vision platform integrating YOLOv8 for real-time object detection with DeepSORT algorithm for multi-object tracking achieving 95%+ ID persistence. Architected Flask REST API backend with 20+ endpoints handling video stream ingestion, processing orchestration, and data persistence via SQLite with optimized indexing. Implemented JWT authentication with role-based access control (admin, operator, viewer roles) and comprehensive audit logging. Developed React 18 + Vite frontend with Tailwind CSS dark theme featuring live MJPEG streaming with bounding box overlays, interactive Leaflet maps for polygonal geofencing, and real-time analytics dashboards with Chart.js visualizations. Engineered multi-threaded video processing pipeline with frame skipping optimization for CPU mode, resolution scaling, and lazy model initialization. Integrated severity-based alerting system with configurable thresholds, time-series analytics, PDF report generation, and CSV/JSON export capabilities. Deployed with Gunicorn WSGI server in cloud-ready configuration (Render/Heroku) with environment-based secrets management.

Key Features

✓YOLOv8 real-time object detection with custom training achieving 50+ mAP50-95 for people, vehicles, drones, and custom object classes
✓DeepSORT multi-object tracking algorithm maintaining 95%+ ID persistence across frames with Kalman filtering and Hungarian assignment
✓Live MJPEG video streaming with real-time bounding box overlays, confidence scores, object classification labels, and track ID visualization
✓Interactive Leaflet map interface for drawing polygonal geofences with real-time breach detection and spatial query optimization
✓Multi-threaded video processing pipeline with frame queue management, GPU/CPU mode switching, and dynamic resolution scaling for performance optimization
✓Severity-based alerting system with configurable thresholds for intrusion detection, loitering analysis, and crowd density monitoring
✓Comprehensive audit logging tracking all user actions, system events, detection triggers, and alert history with timestamp and user attribution
✓Time-series analytics dashboard with object count trends, detection frequency heatmaps, and hourly/daily/weekly aggregation views
✓PDF report generation with executive summary, detection statistics, event timeline, and chart visualizations for security reviews
✓CSV/JSON data export functionality for external analysis, SIEM integration, and historical data archival with filtering options
✓JWT-based authentication with role-based access control supporting admin (full access), operator (monitoring + alerts), and viewer (read-only) roles
✓Secure file upload system with video format validation, async processing queue, and progress tracking for batch analysis
✓Dark theme UI optimized for security operations centers with auto-refreshing components (2-second polling) and minimal latency display
✓Frame skipping optimization for CPU mode reducing computational load while maintaining detection accuracy through intelligent sampling
✓Lazy model initialization and singleton pattern preventing memory duplication and enabling fast server startup times
✓Cloud-ready deployment with Gunicorn WSGI server, environment variable configuration, and compatibility with Render, Heroku, and Railway platforms

Technology Stack

Computer Vision

YOLOv8DeepSORTOpenCVPyTorchNumPy

Backend

FlaskPython 3.13GunicornJWTREST API

Frontend

React 18ViteTailwind CSSLeaflet.jsChart.js

Database & Storage

SQLiteIndexed QueriesJSONFile Upload System

Streaming & Processing

MJPEGMulti-threadingAsync ProcessingFrame Buffering

Deployment

RenderHerokuEnvironment ConfigWSGI Server

Key Outcomes & Impact

  • 1Achieved 50+ mAP50-95 object detection accuracy with custom-trained YOLOv8 model and 95%+ multi-object tracking ID persistence using DeepSORT algorithm
  • 2Delivered real-time video processing at 30+ FPS with GPU acceleration and optimized CPU fallback mode through adaptive frame skipping and resolution scaling
  • 3Implemented production-grade REST API with 20+ endpoints, JWT authentication, role-based access control, and comprehensive audit logging for enterprise security operations
  • 4Built interactive React 18 dashboard with live MJPEG streaming, bounding box overlays, Leaflet-based polygonal geofencing, and real-time analytics visualizations
  • 5Enabled automated threat detection with severity-based alerting, configurable thresholds, and instant notifications for geofence breach and suspicious activity patterns
  • 6Generated PDF/CSV reports with time-series analytics, heatmap visualizations, and historical trend analysis for security audits and operational review

Interested in learning more?

I'd love to discuss this project in detail and share insights about the development process.