🧠 Dynamic Monitoring

📌 Overview

Dynamic Monitoring is an intelligent, multi-camera real-time object detection and tracking system.
It supports dynamic Region of Interest (ROI) setup, multi-threading and multi-processing, and runs on either GPU or CPU — fully configurable via a single configuration file.

The system is designed for scalable surveillance or industrial monitoring setups that require simultaneous camera feeds (video files, RTSP streams, or webcams).

---

🎯 Main Agenda

The main goal of Dynamic Monitoring is to create a fully dynamic and configurable object detection pipeline built on YOLO.
This project allows users to:

• 🧩 Switch YOLO models dynamically — simply update the model path in the config (no code change needed).
• 🎯 Modify or add detection classes directly in config.json.
• 📍 Create custom ROIs per camera interactively to focus detection areas.
• 📊 Get real-time detections and object counts per interval from defined ROIs.
• 🔄 Run detection on multiple cameras simultaneously using multiprocessing or threading.

In short — you can plug in any YOLO model, any class, and any number of cameras, define your ROIs, and get instant detection counts.

---

⚙️ Key Features

• 📸 Multi-camera support — works seamlessly with multiple sources (videos, RTSP, webcam).
• 🚀 Supports both multiprocessing & multithreading — configurable directly from config.json.
• 🧩 Dynamic ROI management — interactive region selection for per-class monitoring.
• 💪 GPU/CPU flexibility — switch inference device anytime without changing code.
• 📊 MySQL database integration — automatic count logging at fixed time intervals.
• ⚡ Frame skip control — allows frame-level optimization for faster inference.
• 🧠 Supports YOLO + ByteTrack tracking pipeline for real-time performance.

---

🧰 Setup & Usage

1️⃣ Create the Virtual Environment

python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

2️⃣ Setup ROIs for Each Camera

python3 main/setup_rois.py

If a ROI file already exists for a camera, you will be prompted to redraw it (y/n). Select the ROI interactively by drawing rectangles on the frame. ROIs are saved automatically in /rois/<model_name>/<camera_name>.json.

3️⃣ Run All Cameras

python3 main/start_all.py

Reads all camera configurations from config.json. Switch between multiprocessing or threading via the mode field in the config. Toggle GPU or CPU usage by setting gpu = true/false in the config. Counts are logged automatically into the database at configured intervals.

4️⃣ Add or Modify Cameras To add a new camera or modify an existing one, update config.json with the new source, name, or type (video, RTSP, webcam). ROIs for new cameras can be created using setup_rois.py.

Folder Overview main/ — core scripts to run cameras, YOLO detection, and ROI setup. rois/ — stores ROI JSON files per camera and model. config/ — contains config.json for camera sources, model path, classes, frame skip, hardware selection, and multiprocessing/threading mode. db/ — database utilities for connection and count logging. roi_handler/ — utilities to interactively select, save, and load ROIs. videos/ — stores video feeds for testing. models/ — YOLO detection models. venv/ — Python virtual environment with dependencies.

📊 Observations

Multiprocessing (GPU)

• GPU utility: 65% average
• Memory usage for 6 cameras (4 videos, 1 RTSP, 1 webcam): 1,846 MiB (~1.8 GB)
• CPU load: average 5.44
• Memory usage: 8.27 GB
• Camera counts per 30-sec interval:
  • 6 cameras (4 videos, 1 RTSP, 1 webcam) → 11 counts
  • 5 cameras (3 videos, 1 RTSP, 1 webcam) → 11 counts
  • 4 cameras (2 videos, 1 RTSP, 1 webcam) → 13 counts
  • 3 cameras (2 videos, 1 RTSP) → 13 counts
  • 3 cameras (2 videos, 1 webcam) → 16 counts
• All 6 cameras run fine on GPU simultaneously.

Threading

• GPU utility: 656 MiB
• CPU load: average 4
• Memory usage: 5.6 GB

Config Flexibility

• Switch between multiprocessing or threading anytime from config.json.
• Switch between GPU or CPU anytime from config.json.