Enterprise-Grade Autonomous Security Monitoring System Based on YOLOv8 and ByteTrack

This project is an enterprise-grade autonomous security monitoring system built with Python, OpenCV, YOLOv8, and ByteTrack. It features functions like polygonal geofencing, dynamic threat scoring, and automatic evidence capture. It can real-time detect intrusion and record evidence in unattended scenarios, suitable for places requiring 24-hour monitoring such as data centers, warehouses, and offices.

With the growing demand for security, traditional monitoring has issues like high cost and manual oversight. Intelligent security based on computer vision and deep learning has become a trend. This project upgrades standard object detection to a complete security solution, enabling real-time video processing, intrusion detection, threat assessment, and evidence preservation without human intervention.

Core Technologies

• YOLOv8: Real-time object detection engine with high speed, high accuracy, multi-scale support, and easy deployment features
• ByteTrack: Multi-object tracking component that recovers occluded targets using low-confidence detection boxes, and achieves high-precision real-time tracking with Kalman filtering + Hungarian algorithm
• OpenCV: Responsible for video reading, preprocessing, visualization, and evidence preservation

System Architecture

Adopts modular design, including main monitoring loop (main.py), configuration management (config.py), detection pipeline (detector.py), tracking integration (tracker.py), geometric calculation (utils/geometry.py), and alert system (utils/alert_system.py) modules.

1. Real-time person detection: Uses YOLOv8n lightweight model to ensure real-time processing of multi-channel video streams
2. Persistent tracking ID: ByteTrack assigns a unique ID to each person to maintain consistency across consecutive frames
3. Polygonal geofencing: Supports arbitrary-shaped restricted areas, using the person's foot position as the reference point to judge intrusion
4. Dynamic threat scoring: Calculates threat value based on intrusion duration, number of intruders, and score decay
5. Three-level threat rating: LOW (green), MEDIUM (yellow), HIGH (red) corresponding to different alert prompts
6. Automatic evidence capture: Automatically saves images with timestamps, intruder IDs, and scores when the threat meets the threshold
7. Off-hours monitoring: Configurable monitoring time periods to avoid false alarms during working hours

Applicable to scenarios such as office security, server room protection, warehouse logistics monitoring, restricted area monitoring, and smart building system integration.

Installation Steps

1. Create a virtual environment
2. Install dependencies: pip install -r requirements.txt
3. Run: python main.py (Pre-trained YOLOv8 weights will be downloaded automatically on first run)

Custom Configuration

• Set parameters like video source, detection threshold, monitoring time period, and polygonal restricted area vertices in config.py
• Adjust parameters such as DETECTION_CONF (detection confidence), THREAT_PER_INTRUSION (threat score per intrusion), EVIDENCE_MIN_INTERVAL_S (evidence capture interval) to adapt to different scenarios

This project's technical highlights include:

1. End-to-end real-time processing flow
2. Intelligent dynamic threat assessment mechanism
3. Flexible and configurable parameter system
4. Modular architecture for easy maintenance and expansion
5. Complete evidence chain to meet compliance requirements

The project provides practical reference for computer vision and security system development.