CCTV-Style Vlog Recorder

Real-time CCTV-style vlog recorder with object detection and face anonymization for NVIDIA Jetson Orin Nano Super.

Features

🎥 Real-time Object Detection - YOLOv8n/YOLO11n for detecting people, objects
😎 Face Anonymization - Multiple glitch effects (mosaic, RGB shift, pixel shuffle)
🚀 30 FPS Performance - Optimized frame-skip detection on Jetson Orin Nano
📹 CCTV Aesthetic - Timestamp, FPS counter, surveillance-style overlay
⚡ Multiple Detection Methods - YOLO-Face, MediaPipe, or Hybrid
🎨 Customizable Effects - 8+ different glitch/mosaic styles
💾 Video Recording - MP4 output with configurable resolution
🖥️ GUI Application - Easy-to-use camera app interface with one-click recording

Hardware Requirements

NVIDIA Jetson Orin Nano Super (tested)
Camera (USB webcam, CSI camera, RealSense, etc.)
JetPack 6.1+

Installation

1. Install Dependencies

pip3 install ultralytics opencv-python mediapipe huggingface_hub pillow

2. Download YOLO Face Detection Model

python3 -c "
from huggingface_hub import hf_hub_download
hf_hub_download(
    repo_id='arnabdhar/YOLOv8-Face-Detection',
    filename='model.pt',
    local_dir='.'
)
mv model.pt yolov8-face.pt
"

3. Clone Repository

git clone <your-repo-url>
cd <repo-directory>
chmod +x vlog_recorder.py

Quick Start

GUI Mode (Recommended - Camera App Style)

Launch the GUI application:

python3 vlog_recorder_gui.py

Or double-click the desktop launcher: CCTV Vlog Recorder

Features:

📹 Real-time preview with all effects applied
⏺️ One-click start/stop recording
📷 Screenshot button
⚙️ Settings panel for camera, model, and effect selection
📊 Live FPS and status display
💾 Auto-generated filenames with timestamp

Controls:

Click "⏺ START RECORDING" to begin recording
Click "⏹ STOP RECORDING" to save video
Click "📷 Screenshot" to capture a frame
Click "⚙ Settings" to adjust camera/detection settings

Command-Line Mode (Advanced)

Basic Usage (Preview Only):

python3 vlog_recorder.py --camera 1

Record with Auto-Generated Filename:

python3 vlog_recorder.py --camera 1 --record

Headless Mode (SSH/Remote):

python3 vlog_recorder.py --headless --camera 1 --duration 60 --output my_vlog.mp4

Performance Optimization

30 FPS Setup (Recommended)

python3 vlog_recorder.py \
  --camera 1 \
  --model yolov8n.pt \
  --resolution 640x480 \
  --face-detector yolo \
  --glitch mosaic_heavy \
  --detection-interval 3 \
  --record

High Quality (13 FPS)

python3 vlog_recorder.py \
  --camera 1 \
  --model yolo11n.pt \
  --resolution 1280x720 \
  --detection-interval 1 \
  --glitch mosaic_shuffle \
  --record

Configuration Options

Camera & Model

Option	Default	Description
`--camera`	0	Camera device ID
`--model`	yolo11n.pt	YOLO model (yolov8n.pt, yolo11n.pt, yolo26n.pt)
`--resolution`	640x480	Camera resolution (320x240, 640x480, 1280x720, 1920x1080)

Detection

Option	Default	Description
`--face-detector`	yolo	Face detection method (yolo, mediapipe, hybrid)
`--confidence`	0.3	Object detection confidence threshold
`--detection-interval`	1	Run detection every N frames (1=every frame, 3=30 FPS boost)

Effects

Option	Default	Description
`--glitch`	combined	Glitch effect type

Available Effects:

rgb_shift - RGB channel shift (cyberpunk style)
pixelate - Standard pixelation
noise - Digital noise
scanlines - VHS scanlines
mosaic_heavy - Large pixel mosaic
pixelate_shuffle - Shuffled pixel blocks
mosaic_shuffle - Heavy mosaic + shuffle (strongest anonymization)
combined - All effects combined

Output

Option	Default	Description
`--record`	False	Enable recording with auto-generated filename
`--output`	None	Custom output video path
`--duration`	None	Recording duration in seconds (headless mode)
`--headless`	False	Run without GUI (for SSH/remote)

Performance Comparison

Resolution Impact (YOLOv8n, detection_interval=1)

Resolution	FPS	Quality	Use Case
320x240	13 FPS	Low	Speed priority
640x480	13 FPS	Medium	⭐ Balanced (recommended)
1280x720	10 FPS	High	Quality priority
1920x1080	~7 FPS	Highest	Max quality

Detection Interval Impact (640x480, YOLOv8n)

Interval	FPS	Tracking Smoothness
1 (every frame)	13 FPS	Perfect
2 (every 2nd)	20 FPS	Excellent
3 (every 3rd)	30 FPS	Good ⭐
5 (every 5th)	35+ FPS	Fair

Model Comparison (640x480, single inference)

Model	Inference Time	Size	Notes
YOLOv8n	40ms	6.2 MB	⭐ Fastest
YOLO11n	43ms	5.4 MB	Balanced
YOLO26n	45ms	5.3 MB	Latest, edge-optimized

Examples

Maximum Speed (30 FPS)

python3 vlog_recorder.py \
  --camera 1 \
  --model yolov8n.pt \
  --resolution 640x480 \
  --detection-interval 3 \
  --glitch mosaic_heavy \
  --record

Maximum Quality

python3 vlog_recorder.py \
  --camera 1 \
  --model yolo11n.pt \
  --resolution 1920x1080 \
  --detection-interval 1 \
  --glitch mosaic_shuffle \
  --record

Lightweight Setup (Low Power)

python3 vlog_recorder.py \
  --camera 1 \
  --model yolov8n.pt \
  --resolution 320x240 \
  --detection-interval 5 \
  --glitch mosaic_heavy \
  --record

Cyberpunk Style

python3 vlog_recorder.py \
  --camera 1 \
  --glitch rgb_shift \
  --record

Keyboard Controls

GUI Application

Use on-screen buttons for all controls
Window close button prompts to save recording if in progress

Command-Line Mode

q - Quit
s - Save screenshot

Face Detection Methods

YOLO-Face (Default, Recommended)

Accurate at all angles
Works with side/downward-facing faces
~50ms inference time
Best for vlogging

MediaPipe

Very accurate for frontal faces
Struggles with side angles
~40ms inference time
Good for static shots

Hybrid

Tries YOLO first, falls back to MediaPipe
Best accuracy but slowest
Use with detection-interval ≥ 3

Project Structure

.
├── vlog_recorder_gui.py      # GUI application (recommended)
├── vlog_recorder.py          # Command-line version
├── yolov8-face.pt            # YOLO face detection model
├── yolo11n.pt                # YOLO object detection model (auto-downloaded)
├── yolov8n.pt                # Alternative YOLO model (auto-downloaded)
├── CCTV_Vlog_Recorder.desktop # Desktop launcher
├── README.md                 # This file
└── vlog_recorder_README.md   # Detailed Korean documentation

Troubleshooting

Camera Not Found

# List available cameras
ls /dev/video*

# Test camera
python3 test_camera.py

Low FPS

Increase detection interval: --detection-interval 3
Lower resolution: --resolution 320x240
Use faster model: --model yolov8n.pt
Simpler effect: --glitch mosaic_heavy

Out of Memory

# Use smaller resolution
python3 vlog_recorder.py --resolution 320x240 --record

No Face Detection

Ensure face is visible and well-lit
Try different detector: --face-detector hybrid
Lower confidence: --confidence 0.2

Performance Tips

30 FPS Goal: Use --detection-interval 3 with 640x480
Best Quality: Use 1280x720 with --detection-interval 1
Lowest Latency: Use 320x240 with --detection-interval 5
Battery Saving: Use 320x240 with YOLOv8n

Credits

YOLO: Ultralytics
YOLO-Face: arnabdhar/YOLOv8-Face-Detection
MediaPipe: Google MediaPipe

License

MIT License - See LICENSE file for details

Contributing

Contributions welcome! Please feel free to submit a Pull Request.

Changelog

v1.1.0 (2026-03-15)

NEW: GUI application with camera app-style interface
NEW: One-click start/stop recording
NEW: Real-time preview with settings panel
NEW: Desktop launcher for easy access
IMPROVED: Live FPS and status monitoring
IMPROVED: Auto-generated filenames
IMPROVED: Screenshot functionality in GUI

v1.0.0 (2026-03-15)

Initial release
YOLO object detection
YOLO-Face & MediaPipe face detection
8 glitch effects
Frame-skip optimization (30 FPS)
Headless mode support
Configurable resolution

Optimized for NVIDIA Jetson Orin Nano Super 🚀

Name		Name	Last commit message	Last commit date
Latest commit History 3 Commits
.gitignore		.gitignore
README.md		README.md
vlog_recorder.py		vlog_recorder.py
vlog_recorder_README.md		vlog_recorder_README.md
vlog_recorder_gui.py		vlog_recorder_gui.py

Folders and files

Latest commit

History

Repository files navigation