How to Configure System Settings Using config.yaml in Multi-Cam Face Tracker

To configure system settings in multi-cam-face-tracker, edit the config/config.yaml file and restart the application; the load_config() function in main.py parses the YAML and initializes all runtime parameters as a Python dictionary.

The aarambhdevhub/multi-cam-face-tracker repository centralizes all runtime behavior in a single YAML configuration file. When you configure the system settings using config.yaml, you control everything from face detection thresholds and GPU acceleration to Telegram alert routing without modifying source code.

Where Configuration Lives: The config.yaml File Structure

The application expects its configuration at config/config.yaml. During startup, main.py invokes load_config('config/config.yaml') to ingest settings. This function uses yaml.safe_load to parse the file into a Python dictionary, creates required directories for screenshots, known faces, and logs, then passes the configuration object to the MainWindow, logger, camera manager, and Telegram notifier.

Core Configuration Sections

The YAML file organizes settings into three top-level keys: app, recognition, and telegram.

Application Settings (app)

The app section defines global application behavior and file system paths:

• name — Human-readable application identifier (e.g., "Multi-Cam Face Tracker").
• version — Current release version (e.g., "1.0.0").
• threshold — Global face-match confidence threshold (e.g., 0.6).
• screenshot_dir — Target directory for captured screenshots (e.g., "data/screenshots").
• known_faces_dir — Directory containing reference images for known faces (e.g., "data/known_faces").
• database_path — SQLite database file for detection logging (e.g., "data/database.db").
• alert_sound — Path to the audio file played during alerts (e.g., "assets/alert.wav").
• logo — Path to the splash-screen logo (e.g., "assets/logo.png").
• log_dir — Directory for rotating log files (e.g., "logs").

Face Recognition Parameters (recognition)

The recognition section controls the deep learning pipeline and hardware acceleration:

• detection_threshold — Minimum confidence for the detector model (e.g., 0.5).
• recognition_threshold — Minimum confidence for the recognizer model (e.g., 0.6).
• max_batch_size — Number of faces processed simultaneously for GPU/CPU optimization (e.g., 8).
• device — Compute device selection: "cpu" or "cuda" for GPU acceleration.
• analysis_enabled — Master toggle for age/gender/emotion analysis (true or false).
• age_estimation — Specific toggle for age estimation.
• gender_detection — Specific toggle for gender detection.

Telegram Alert Integration (telegram)

The telegram section configures remote alerting via the Telegram Bot API:

• enabled — Master switch for Telegram notifications (true or false).
• bot_token — Authentication token obtained from @BotFather (keep secret).
• chat_id — Destination chat ID for alert messages.
• rate_limit — Minimum seconds between successive messages to prevent spam (e.g., 30).

How Configuration Loading Works

The load_config function in main.py orchestrates the initialization sequence. It validates file existence, parses YAML safely, and ensures the file system structure matches the configuration.

from pathlib import Path
import yaml
from loguru import logger

def load_config(config_path: str) -> dict:
    with open(config_path, "r") as f:
        config = yaml.safe_load(f)

    # Ensure required directories exist

    Path(config["app"]["screenshot_dir"]).mkdir(parents=True, exist_ok=True)
    Path(config["app"]["known_faces_dir"]).mkdir(parents=True, exist_ok=True)
    Path(config["app"]["log_dir"]).mkdir(parents=True, exist_ok=True)

    return config

After loading, the dictionary propagates through the application: MainWindow receives it for UI configuration, the camera manager extracts device and threshold settings, and the Telegram notifier reads bot credentials.

Practical Configuration Examples

Adjusting Detection Sensitivity

To reduce false positives by requiring higher confidence scores, modify the recognition section:

recognition:
  detection_threshold: 0.7   # increased from 0.5 for stricter detection

  recognition_threshold: 0.6
  max_batch_size: 8
  device: "cpu"
  analysis_enabled: true
  age_estimation: true
  gender_detection: true

Disabling Telegram Notifications

To stop alert messages without removing your bot configuration:

telegram:
  enabled: false
  bot_token: "YOUR_BOT_TOKEN"
  chat_id: "YOUR_CHAT_ID"
  rate_limit: 30

Switching Between CPU and GPU

For CUDA-enabled systems, change the compute device to accelerate face recognition:

recognition:
  detection_threshold: 0.5
  recognition_threshold: 0.6
  max_batch_size: 8
  device: "cuda"   # switch from "cpu" to use GPU acceleration

  analysis_enabled: true

Summary

• Centralized configuration: All runtime parameters live in config/config.yaml, parsed by load_config() in main.py using yaml.safe_load.
• Three main sections: app (paths and global settings), recognition (ML pipeline and hardware), and telegram (alert integration).
• Automatic directory setup: The loader creates screenshot, known-face, and log directories automatically if they do not exist.
• Restart required: Changes to config.yaml only take effect after restarting the application.

Frequently Asked Questions

Where is the config.yaml file located?

The configuration file is located at config/config.yaml relative to the project root. The main.py script specifically calls load_config('config/config.yaml') during startup to locate and parse this file.

Do I need to restart the application after editing config.yaml?

Yes, you must restart the program for changes to take effect. The load_config function reads the YAML file only once during application initialization; it does not watch for file changes or hot-reload configurations while running.

How do I switch from CPU to GPU processing?

Change the device key in the recognition section from "cpu" to "cuda". Ensure your system has CUDA-capable hardware and the appropriate PyTorch CUDA drivers installed. You may also want to increase max_batch_size to utilize GPU memory efficiently.

Can I disable specific analysis features like age estimation?

Yes, set analysis_enabled to false to disable all demographic analysis, or selectively toggle individual features by setting age_estimation, gender_detection, or related keys to false under the recognition section. This reduces computational overhead if you only need basic face detection and recognition.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}