AI CCTV Detection Extension (Local Surveillance Prototype)

Overview

AI-powered local surveillance system built on top of an existing CCTV stream accessed through Bluestacks + ezyKam+.

The system captures CCTV frames locally, performs real-time computer vision analysis, identifies suspicious activity, and stores evidence without requiring cloud-based CCTV subscriptions.

---

High-Level Pipeline

CCTV Camera
      ↓
ezyKam+
      ↓
Bluestacks
      ↓
Background Frame Capture
      ↓
Person Detection (YOLO)
      ↓
Person Tracking (ByteTrack)
      ↓
Zone Mapping / Virtual Fence
      ↓
Face Recognition
      ↓
Risk Assessment Engine
      ↓
Evidence Generation
      ↓
Local Storage & Alerts

---

Architecture

Capture
↓
Detection
↓
Tracking
↓
Zone Mapping
↓
Face Recognition
↓
Risk Engine
↓
Recording & Alerts

---

Core Concepts

Background Window Capture

Capture CCTV frames directly from Bluestacks even when the application is running in the background.

Person Detection

Detect human presence in CCTV footage using a lightweight YOLO model optimized for CPU inference.

Multi-Object Tracking

Assign persistent IDs to detected individuals and track their movement across frames using ByteTrack.

Face Recognition

Identify known individuals and distinguish them from unknown visitors using facial embeddings.

Zone Mapping

Define custom zones such as:

• Public Area
• Entry Gate
• Boundary Wall
• Restricted Area
• House Entrance

and monitor movement between them.

Virtual Fence Detection

Detect intrusion events such as:

• Crossing restricted boundaries
• Entering prohibited zones
• Jumping walls
• Unauthorized access attempts

Risk Engine

Generate risk scores based on multiple factors:

• Known vs Unknown Person
• Restricted Area Access
• Time of Day
• Face Visibility
• Movement Pattern
• Zone Violations

Event-Based Recording

Automatically save evidence clips only when meaningful events occur.

Local Evidence Storage

Store:

• Snapshots
• Detection Events
• Annotated Videos
• Face Data
• Tracking Information

locally on the machine.

---

Planned Features

• Human Detection
• Multi-Person Tracking
• Known vs Unknown Face Recognition
• Visitor Logging
• Virtual Fence Detection
• Zone-Based Monitoring
• Suspicious Activity Detection
• Event-Based Video Recording
• Evidence Management
• Local Database Integration
• Alert System
• Dashboard & Analytics
• Adaptive Learning Enhancements (Future)

---

Technologies

Computer Vision

• OpenCV
• YOLOv8
• ByteTrack
• InsightFace / ArcFace
• NumPy

Capture Layer

• pywin32
• Windows PrintWindow API
• Bluestacks
• ezyKam+

Storage

• MongoDB
• MongoDB Compass
• Local Filesystem Storage

Backend

• Python

Future Extensions

• Reinforcement Learning
• Risk Scoring Engine
• Notification Services
• Web Dashboard

---

Project Goals

• Eliminate dependency on cloud CCTV subscriptions
• Perform all processing locally
• Detect and track people in real time
• Identify unknown visitors
• Detect intrusion events
• Preserve only relevant evidence
• Build an intelligent surveillance system capable of understanding activity rather than simply recording footage