DVR-YOLOv8-Detection

Description

dvr-yolov8-detection is designed for real-time detection of humans, animals, or objects using the YOLOv8 model and OpenCV.

It supports real-time video streams via RTMP or USB webcams, includes CUDA GPU acceleration for enhanced performance, and provides options for saving detections, triggering alerts and logging events.

The video preview can be run both in a GUI window and headless on a local web server using the included Flask web server setup.

(New) 🐳 A Dockerfile for Dockerized installs is also included.

Features

• Real-time human/animal/object detection and alert system
• Runs on Python + YOLOv8 + OpenCV2
• CUDA GPU acceleration; also runs in CPU-only mode
• Supports either RTMP streams or USB webcams for real-time video sources
  • Includes a loopback example and NGINX configuration example for RTMP use (i.e. OBS Studio)
• Detections can be automatically saved as images with a detection log
• Separate tool included for offline video file detections (see: utils/)

Overview

The project uses Python with YOLOv8 and OpenCV2 to perform object detection on either real-time video streams or batches of video files. It processes each video frame by frame, detecting humans by default (other YOLOv8-supported objects can be added as needed). The program can log detections into a separate log file, save detection frames with highlighted objects, and send out audio alerts via pyttsx3.

Configuration options are available in the config.ini file.

Note: Using CUDA-enabled OpenCV is recommended for faster operation. CUDA-enabled OpenCV2 needs to be compiled manually and installed separately, as the compiling process is highly dependent on your hardware setup. Refer to the "Troubleshooting" section for guidance and an example build script for OpenCV with CUDA.

Real-time detection also supports additional CUDA features such as CUDA video denoising (note: this feature requires CUDA and is often available only when OpenCV is compiled from source).

Configurable Features via config.ini:

• Video source (USB webcam or RTMP stream)
• Confidence threshold for detections
• Enable or disable rescaling of video frames
• CUDA-based video denoising (experimental)
• Log detection details into a separate log file
• Save frames with detected objects as image files
• Model variant selection (e.g., YOLOv8n, YOLOv8s, YOLOv8m)
• ... and other additional customizable options

---

🐳 For Docker setup, see the DOCKER_SETUP.md for a guide.

---

Setup

Requirements

• Python 3.6+ (Python 3.10.x recommended)
  • Python modules:
  • See requirements.txt
• FFmpeg
• Python 3.10.x
• If you wish to use CUDA GPU acceleration, you will need:
  • A Nvidia GPU that supports CUDA
  • Install CUDA 11.8 or higher to enable GPU-accelerated processing
• Use Miniconda or Mamba for environment management

Installation (Conda/Mamba Environments)

1. Clone the repository:

   git clone https://github.com/FlyingFathead/dvr-yolov8-detection.git
   cd dvr-yolov8-detection

1.2 (Install Miniconda or Anaconda if not already installed):

• Download and install Miniconda (recommended):

  • For Linux/macOS:

    wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
    bash Miniconda3-latest-Linux-x86_64.sh

  • For Windows, download the installer from here and follow the installation instructions.

2. Set up the environment Conda/Mamba environment:

   ./setup_mamba_env.sh

   This script creates a Conda/Mamba environment with the required dependencies.

3. Run the detection script:

   ./run_detection.sh

Installation (Manual Steps)

1. Clone the repository:

   git clone https://github.com/FlyingFathead/dvr-yolov8-detection.git
   cd dvr-yolov8-detection

2. Install the required Python packages:

   pip install -r requirements.txt

   This ensures that all dependencies are managed through the requirements.txt file.

3. Install OpenCV:

   • For a full version with GUI support (recommended):

     pip install opencv-python

   • For CUDA-enabled OpenCV, you need to build it from source. Refer to the OpenCV documentation or check the build script for guidance.

4. Install FFmpeg:

   • On Ubuntu:

     sudo apt-get install ffmpeg

   • On Windows and macOS, follow the instructions on the FFmpeg download page.

Real-Time Detection

This project supports real-time object detection from RTMP streams or USB webcams using YOLOv8. The provided run_detection.sh script ensures the detection script runs continuously, automatically restarting if it exits.

(New in v0.155): The real-time detection now comes with a mini web server running on Flask that enables you to run the detection framework in your browser by default when headless and enable_webserver are set to true. This will make headless deployment very easy. Note that it listens on 0.0.0.0:5000 by default, which you might want to change (i.e. to 127.0.0.1:5000) for safety reasons.

Usage

1. For RTMP Streams:

• Set up your RTMP server:

  • Use the example NGINX configuration (example-nginx.conf) to set up an RTMP loopback server.
  • Ensure your streaming client (e.g., OBS Studio) is set to stream to rtmp://127.0.0.1:1935/live.

• Run the detection script:

  ./run_detection.sh

2. For USB Webcams:

• Run the detection script with the --use_webcam option:

  python3 yolov8_live_rtmp_stream_detection.py --use_webcam true

• Specify webcam index (if needed):

  python3 yolov8_live_rtmp_stream_detection.py --use_webcam true --webcam_index <number>

  • Replace <number> with the index number of your webcam.
  • Use the utils/get_webcams.py tool to find available webcams and their index numbers.

• Alternatively, configure via config.ini:

  • Set use_webcam to true.
  • Set webcam_index to your desired webcam index.

Configuration

Running Headless / As A Web Server

Due to Docker being a popular install option, you can run the program headless and with a Flask based mini-web server included. Compared to the regular GUI verison, there is likely a small framerate dip and latency in the output, but other than that, the functionality is not too far off from the GUI variant.

In headless Docker installs, make sure that headless and enable_webserver are both set to true.

Editing Program Configuration

You can configure the program's parameters by editing the config.ini file. This allows you to set program parameters such as the input source, input stream address, output directory, confidence threshold, model variant, stream URL, and more.

(For RTMP Sources) Example NGINX Configuration

An example NGINX configuration is provided in example-nginx.conf. This config sets up an RTMP server that listens on 127.0.0.1:1935 and allows local clients to publish and play streams.

(For RTMP Sources) RTMP Loopback Script

To stream and process the video in real-time, use the ffmpeg_video_stream_loopback.sh script. Ensure your streaming client (e.g., OBS Studio) is set to stream to rtmp://127.0.0.1:1935/live.

(For RTMP Sources) Windows Users / Platform-Agnostic Loopback for RTMP

Use the utils/loopback_test_unit_ffmpeg-python.py script to set up a loopback for your RTMP stream.

1. Install ffmpeg-python:

   pip install -U ffmpeg-python

2. Run the loopback script:

   python3 utils/loopback_test_unit_ffmpeg-python.py

3. Run the detection script:

   python3 yolov8_live_rtmp_stream_detection.py

4. Configure your streaming client to stream to:

   rtmp://127.0.0.1:1935/live
   

Note: Using NGINX as a loopback method is highly recommended for stability.

Offline Batch Detection Utility

Use utils/batch_humdet_yolo8_opencv2.py to run YOLOv8 batch detection on directories of video files, suitable for faster offline use.

Troubleshooting

RTMP Loopback Issues

• If for whatever reason the loopback doesn't seem working, you can create a test stream with i.e. utils/test_stream_generator.py. When ran, the script generates synthetic video frames and streams them to your RTMP server using FFmpeg as a subprocess, enabling you to try out if your loopback works.
• Run the test_stream_generator.py and keep it running in the background, then try to first use VLC to connect to your stream (VLC: Media -> Open Network Stream -> rtmp://127.0.0.1:1935/live/stream). If this works, the main detection script should work as well.

CUDA Not Found

• Ensure that you have all necessary modules installed with CUDA enabled.
• You may need to compile OpenCV from source to enable CUDA support.
• Refer i.e. to the OpenCV w/ CUDA build script for Ubuntu 22.04LTS or the 24.04LTS build script for some degree of guidance.
• Verify CUDA support by checking if the program detects your GPU on startup.

CUDA Installation

• See: CUDA Setup Guide

Running Without a CUDA-Capable GPU

• The program can run in CPU-only mode, though performance may be slower.
• To improve performance:
  • Use a smaller model size in the config.ini.
  • Adjust rescale and frame rate options.
  • Reduce the resolution and frame rate from the video source.

TODO

• Implement more error handling for edge cases
• Refactor code for improved modularity
• Add setup scripts for easier deployment
• Implement threshold settings for alerts (e.g., number of detections within a time frame)
• Add hooks for sending detections to web servers or APIs

Changelog

• v0.1607
  • New: Get detection alerts via Telegram (optional)
  • Use @BotFather on Telegram to create a bot token
  • Set your userid(s) (can be multiple users, comma separated) and the bot API token as environment variables:
    • DVR_YOLOV8_ALLOWED_TELEGRAM_USERS
      • allowed users/send alerts to (comma-separated list)
    • DVR_YOLOV8_TELEGRAM_BOT_TOKEN
      • your Telegram bot API token for alerts
• v0.1606
  • Performance improvements:
    • switched to PyAV-based handling for better RTMP stream reliability with less CPU load
  • UI/UX improvements:
    • Web server UI/UX enhancements in image carousel browsing
    • shows detection area by default, swappable between full frame/area
    • clicking on the image now shows its original version
    • better error/exception catching overall
    • better webUI scaling in various devices etc
• v0.1605 Overall compatibility & bug fixes
  • Detection image carousel beta over webUI
  • If detection saving is enabled, images can be viewed from webUI
• v0.1604 Frame queue sizes now configurable
  • Helps in I/O performance issues when saving detections
  • config.ini => [performance] => frame_queue_size
• v0.1603 New configuration for saving
  • Choose to save the detection area, the whole frame, or both.
  • (see config.ini => save_full_frames and save_detection_areas)
• v0.1602 Queuing on image saving
  • Should reduce lag on most systems, even with larger frames
• v0.1601 Active access logging for webUI connections; improved
  • Access via webUI is logged by default to logs/access.log
  • See config.ini for more options
• v0.160 (Oct-13-2024) WebUI access logging added
  • Can be enabled/disabled and defined in config.ini
• v0.159 (Oct-12-2024) Fixes to the detection saving logic
• v0.158 (Oct-11-2024) Even more webUI updates
  • Human detections get aggregated in the webUI within a cooldown period
  • (default cooldown period: 30 seconds)
• v0.157 (Oct-11-2024) webUI updates
  • Better refreshed data via AJAX
  • Minimized lock holding time web_server.py => better FPS
  • webserver_max_fps value to cap the framerate on the webUI for better performance
• v0.156 (Oct-11-2024) Detection graphs in web UI
  • Added matplotlib based detection graphs to the web UI
  • (selectable between 1hr/24hrs/week/month/year)
• v0.155 (Oct-11-2024) Now comes with a Flask web server!
  • The video feed can be monitored real-time using the web interface
  • Added a Flask mini web server to take care of the streams
  • enable_webserver and headless both set to true by default
  • Server listens at 0.0.0.0:5000 (see config.ini for more)
  • This enables quick deployment especially in headless / Docker setups
• v0.154 (Oct-10-2024) 🐳 Dockerized Setup Now Available! 🐳
  • Headless mode added for non-GUI/Docker/detection-only modes
    • enable in config.ini with headless = true
    • or, run the program with --headless flag
  • Added Docker as an install method to ease the setup process
  • additional installation guides
• v0.153
  • config.ini & program changes:
  • Fallback directory (fallback_save_dir)
  • Option to create date-based sub-directories (i.e. /yolo_detections_path/year/month/day/)
• v0.152
  • Added Conda/Mamba installer script for easier deployment
• v0.151
  • Added fallbacks for directories
• v0.1501
  • Fallback to non-CUDA modes if CUDA not supported
• v0.15
  • Added direct USB webcam support
    • Configure via config.ini or use --use_webcam flag
    • Added utils/get_webcams.py to find webcam indices
• v0.1402
  • Added detection logging to a file
• v0.1401
  • Added configparser; now configurable via config.ini
• v0.140
  • Improved configuration option checks

Licensing

This project is licensed under the GNU General Public License v3.0 (GPLv3).

Ethos and Intent

I created this project to support non-profit and educational endeavors. While the GPLv3 license permits commercial use, I kindly ask that if you plan to use this project for commercial purposes, you consider reaching out to me. Your support and collaboration are greatly appreciated.

Contact

For inquiries, suggestions, or collaborations, please contact me at flyingfathead@protonmail.com or visit FlyingFathead on GitHub.

Contributing

Contributions are welcome! Feel free to open issues or submit pull requests on GitHub, or contact the author (me) directly at flyingfathead@protonmail.com.

Credits

Developed by FlyingFathead, with digital ghost code contributions from ChaosWhisperer.

Other

Star it if you like it. *;-)