IPyCam - Pure Python Virtual IP Camera

A lightweight, pure Python virtual IP camera that provides ONVIF discovery, RTSP streaming, and PTZ controls. Perfect for testing, development, or creating custom camera solutions.

📺 Demo Video

Untitled720p.mp4

YouTube

Features

• ONVIF Compliance: Full WS-Discovery support for automatic camera detection
• RTSP Streaming: High-performance video streaming via go2rtc (optional) or native Python fallback
• WebRTC Support: Direct browser streaming with native Python implementation (optional extra)
• PTZ Controls: Digital Pan-Tilt-Zoom with preset positions
• Hardware Acceleration: Automatic detection and use of NVENC, QSV, or CPU encoding (with go2rtc)
• Native Fallback: Pure Python streaming (WebRTC, MJPEG) when go2rtc is not available
• Optional Authentication: HTTP Basic auth for the web UI/REST API and WS-Security for ONVIF, off by default
• Local Recording: record the stream to disk (mp4/avi) with an optional pre-record buffer
• Display Transforms: flip, mirror, and 90/180/270 rotation, applied live
• Web Interface: Built-in configuration (tabbed settings), live preview with main/sub stream switching, and recording control
• Decoupled Frame Pipeline: capture never blocks on encoding, disk I/O, or slow clients (bounded drop-oldest queues + per-output worker threads)
• Silent-by-Default Logging: the library emits nothing unless the host application opts in
• Flexible Input: Accept frames from any source (webcam, video file, generated content)

Requirements

• Python 3.8+
• Optional: FFmpeg + go2rtc for hardware-accelerated encoding (recommended for high performance)
• Optional: pip install ipycam[webrtc] (pulls in aiortc + aiohttp) for native Python WebRTC streaming

Note: IPyCam can run without go2rtc using pure Python streaming. However, go2rtc + FFmpeg provides significantly better performance, especially for high-resolution streams.

Note: aiortc/aiohttp are no longer installed by default -- the core package installs without them. Install the webrtc extra only if you need the native (non-browser, non-go2rtc) WebRTC path.

Quick Start

Clone the repo:

git clone https://github.com/olkham/IPyCam.git

Setup Scripts (Windows and Linux)

If you prefer a guided setup, use the provided scripts. They install dependencies and prepare the environment for running the examples.

Windows (PowerShell or Command Prompt):

setup.bat

Linux/macOS (bash):

chmod +x setup.sh
./setup.sh

Tip: Run the script from the project root (the folder that contains setup.bat and setup.sh).

Installation

Install directly from GitHub:

pip install git+https://github.com/olkham/IPyCam.git

Or install from source:

git clone https://github.com/olkham/IPyCam.git
cd ipycam
pip install -e .

Optional: Enhanced Streaming Performance (Recommended)

For hardware-accelerated encoding with go2rtc:

1. Install go2rtc: Download from go2rtc releases
2. Start go2rtc with IPyCam configuration:

   go2rtc.exe --config ipycam\go2rtc.yaml

   Keep this running in a separate terminal.

Without go2rtc, IPyCam will automatically fall back to native Python streaming.

Optional: Native WebRTC Support

For the pure-Python native WebRTC streaming path (used when go2rtc isn't running and you still want direct browser streaming):

pip install "ipycam[webrtc]"

This pulls in aiortc and aiohttp. Without this extra, IPyCam still works fine via go2rtc/RTSP/MJPEG -- native WebRTC is simply unavailable and the web UI falls back to MJPEG preview.

Optional: 360° Camera Support

For the 360_ptz.py example with equirectangular projection:

pip install "ipycam[camera360]"

or install FrameSource separately:

pip install framesource

Basic Usage

import cv2
from ipycam import IPCamera, CameraConfig

# Create camera with custom config
config = CameraConfig(
    name="My Virtual Camera",
    main_width=1920,
    main_height=1080,
    main_fps=30,
)

camera = IPCamera(config)
camera.start()

# Stream from webcam
cap = cv2.VideoCapture(0)

try:
    while camera.is_running:
        ret, frame = cap.read()
        if ret:
            camera.stream(frame)
except KeyboardInterrupt:
    pass
finally:
    cap.release()
    camera.stop()

Running as a Module

python -m ipycam

The library itself is silent by default (see Logging); the CLI enables console logging for you and supports --log-level {DEBUG,INFO,WARNING,ERROR} (default INFO) to control verbosity, e.g.:

python -m ipycam --log-level DEBUG

python -m ipycam --config custom.json loads (and, on any config change made via the web UI, saves back to) that specific file instead of the default camera_config.json.

Then access:

• Web UI: http://localhost:8080/
• RTSP Main Stream: rtsp://localhost:8554/video_main
• RTSP Sub Stream: rtsp://localhost:8554/video_sub
• ONVIF Service: http://localhost:8080/onvif/device_service

Testing the Stream

Test the RTSP streams using ffplay:

# Test main stream
ffplay rtsp://localhost:8554/video_main

# Test sub stream
ffplay rtsp://localhost:8554/video_sub

Configuration

Configuration is stored in camera_config.json (or whatever path is passed to --config / CameraConfig.load() -- changes made via the web UI are saved back to that same file):

{
  "name": "Virtual Camera",
  "manufacturer": "PythonCam",
  "model": "VirtualCam-1",
  "username": "",
  "password": "",
  "main_width": 1920,
  "main_height": 1080,
  "main_fps": 30,
  "main_bitrate": "8M",
  "sub_width": 640,
  "sub_height": 360,
  "sub_bitrate": "1M",
  "hw_accel": "auto",
  "flip": false,
  "mirror": false,
  "rotation": 0,
  "recording_enabled": false,
  "recording_format": "mp4",
  "recording_path": "recordings",
  "recording_max_file_mb": 1024,
  "recording_pre_seconds": 0
}

Hardware acceleration options (go2rtc only):

• "auto" - Try NVENC → QSV → CPU (default)
• "nvenc" - NVIDIA GPU encoding
• "qsv" - Intel Quick Sync Video
• "cpu" - Software encoding (libx264)

Display transforms (flip/mirror/rotation) are applied to every outbound frame (after PTZ, before the timestamp overlay) regardless of streaming mode:

• flip - vertical flip (upside-down)
• mirror - horizontal flip (mirror image)
• rotation - clockwise rotation in degrees: 0, 90, 180, or 270

Recording fields are covered in Recording below. username/password are covered in Authentication.

Authentication

Authentication is optional and off by default. With empty username/password, every endpoint behaves exactly as before (fully open).

To enable it, set both a username and password -- either:

• In the web UI, under the User settings tab, or
• Via POST /api/credentials with a JSON body {"username": "...", "password": "..."}, or
• Directly in the config file (username/password fields).

Once both are set:

• HTTP Basic auth protects the web UI, the REST API (/api/*), the snapshot endpoint, and the MJPEG stream.
• ONVIF requests must carry a WS-Security UsernameToken (PasswordDigest); unauthenticated SOAP requests get a fault response.
• /static/* assets remain unauthenticated (CSS/JS/images only -- no camera data).

Setting both fields back to empty strings disables auth again.

Security note: IPyCam binds to 0.0.0.0 by default (reachable from any device on your network). If you're running it on an untrusted network, enabling authentication is recommended.

Logging

The ipycam library is silent by default -- it attaches only a NullHandler to its logger tree, so importing/using it produces no console output. To see log messages, either:

from ipycam import configure_logging
import logging

configure_logging(level=logging.INFO)  # or DEBUG for more detail

or configure the "ipycam" logger yourself with the standard logging module. python -m ipycam calls configure_logging() for you and exposes it via --log-level.

Web Interface

The built-in web UI (http://<host>:8080/ by default) provides:

• A live preview with a stream switcher (go2rtc RTC main/sub, native WebRTC, or native MJPEG) and, in native MJPEG mode, a main/sub quality toggle
• A record button that starts/stops local recording via the same API described in Recording
• Tabbed settings: Display (timestamp, flip/mirror/rotation), Stream (main/sub resolution, FPS, bitrate, hardware acceleration, read-only network ports), Identity (name/manufacturer/model), and User (authentication credentials)
• Each settings tab saves independently; fields that require a stream restart (e.g. resolution, FPS, rotation) trigger one automatically

Recording

IPyCam can record the outbound stream to disk independently of any RTSP/WebRTC/MJPEG client. Recording runs on its own worker thread, so a slow disk drops frames from the recorder's internal queue rather than affecting the live stream.

Configure via camera_config.json:

• recording_enabled - if true, the recorder (and its pre-record buffer, if configured) starts running as soon as the camera starts
• recording_format - "mp4" or "avi"
• recording_path - output directory (config-file only; not editable via the web UI/API, to avoid a path-traversal foot-gun)
• recording_max_file_mb - segment rotates to a new file once it exceeds this size
• recording_pre_seconds - seconds of pre-record buffer (0-30) to include before the trigger

Control at runtime via the web UI Record button, or the REST API:

curl -X POST http://localhost:8080/api/recording/start
curl -X POST http://localhost:8080/api/recording/stop
curl http://localhost:8080/api/recording/status

or directly from Python:

camera.start_recording()
# ... later ...
files = camera.stop_recording()  # returns the finalized segment file path(s)

Memory caveat: the pre-record buffer holds recording_pre_seconds * main_fps full decoded frames at main resolution in memory. At high resolution/FPS this can be large (e.g. ~5 seconds at 30fps/1080p is several GB), which is why recording_pre_seconds is capped at 30. Keep it as low as your use case allows.

PTZ Controls

The camera includes digital PTZ (Pan-Tilt-Zoom) support:

# Access PTZ through ONVIF or directly
camera.ptz.continuous_move(pan_speed=0.5, tilt_speed=0.0, zoom_speed=0.0)
camera.ptz.stop()
camera.ptz.goto_preset(preset_token="preset1")

PTZ presets are stored in ptz_presets.json.

Performance Tips

1. Use hardware acceleration: Enable NVENC (NVIDIA) or QSV (Intel) for best performance
2. Match resolutions: Set camera input to match streaming resolution to avoid resize overhead
3. Adjust FPS: Most webcams are limited to 30fps
4. Disable PTZ: Set camera to home position (0,0,0) to skip PTZ transforms
5. The capture loop never blocks on output: MJPEG encoding, native RTSP fan-out, recording, and go2rtc/WebRTC delivery all run on their own worker threads behind bounded drop-oldest queues, so a slow client, a slow disk, or a stalled encoder degrades only that consumer (extra frame drops there) rather than the whole camera
6. A dead go2rtc/FFmpeg process auto-recovers: IPyCam retries with a bounded backoff and, if reconnection is exhausted, falls back to serving MJPEG/snapshot instead of taking the camera down

Architecture

ipycam/
├── __init__.py          # Package exports
├── __main__.py          # CLI entry point
├── camera.py            # Main IPCamera class (frame pipeline orchestration)
├── config.py            # CameraConfig dataclass
├── streamer.py          # go2rtc/FFmpeg encoding and streaming pipeline
├── ptz.py               # Digital PTZ implementation
├── onvif.py             # ONVIF SOAP service (incl. WS-Security auth)
├── discovery.py         # WS-Discovery server
├── http.py              # HTTP request handler (web UI, REST API, auth)
├── mjpeg.py             # Native MJPEG streaming fallback
├── rtsp.py              # Native RTSP server fallback
├── webrtc.py            # Native WebRTC streaming (optional `webrtc` extra)
├── framequeue.py        # Bounded drop-oldest frame queue (pipeline primitive)
├── recorder.py          # Local disk recording (segments + pre-record buffer)
├── logging_config.py    # configure_logging() helper (library is silent by default)
└── static/              # Web UI and SOAP templates

Development

Running Tests

IPyCam includes a comprehensive test suite (400+ tests) using pytest. To run the tests:

# Install dev dependencies (add the webrtc extra too for full WebRTC coverage)
pip install -e ".[dev,webrtc]"

# Run all tests
pytest

# Run with verbose output
pytest -v

# Run with coverage report
pytest --cov=ipycam --cov-report=term-missing

# Run specific test file
pytest tests/test_config.py
pytest tests/test_ptz.py
pytest tests/test_mjpeg.py
pytest tests/test_onvif.py

Tests that exercise the webrtc extra are skipped automatically when aiortc isn't installed, so pip install -e ".[dev]" alone is sufficient for most changes. See CONTRIBUTING.md for more on the dev workflow, linting/type-checking, and CI.

The test suite covers:

• CameraConfig: Configuration serialization, URL generation, hardware acceleration settings, auth/display/recording field validation
• PTZController: Positioning, presets, hardware handler callbacks, frame transforms
• MJPEGStreamer: Client management, frame streaming, statistics, main/sub stream selection
• ONVIFService: SOAP response generation, PTZ command parsing, device info, WS-Security UsernameToken auth
• IPCameraHTTPHandler (http): REST/config endpoints, HTTP Basic auth, static-file path-traversal hardening, recording endpoints, video upload limits
• IPCamera (camera): frame pipeline orchestration, streaming-mode selection/fallback, recording integration
• VideoStreamer (streamer): go2rtc/FFmpeg lifecycle, reconnect/backoff behavior
• NativeRTSPServer (rtsp): native RTSP fallback streaming
• NativeWebRTCStreamer (webrtc): native WebRTC signaling and streaming (skipped without the webrtc extra)
• WSDiscoveryServer (discovery): WS-Discovery responses and lifecycle
• FrameQueue (framequeue): bounded drop-oldest queue semantics
• VideoRecorder (recorder): segment rotation, pre-record ring buffer, start/stop lifecycle

CI (.github/workflows/ci.yml) runs this suite on ubuntu-latest and windows-latest across Python 3.8-3.12 on every push to main and every pull request, plus non-blocking ruff/mypy checks.

Troubleshooting

Camera freezes after a few frames

• With go2rtc: Check FFmpeg is installed and in PATH, verify go2rtc.exe is running. A dead FFmpeg process is retried automatically (bounded backoff); if it can't recover, the camera falls back to MJPEG-only rather than freezing -- look for a "falling back to MJPEG-only streaming" warning in the log
• Native mode: Install PyAV (pip install av) for RTSP or pip install "ipycam[webrtc]" for native WebRTC
• Check hardware encoder availability (go2rtc only)

Low FPS performance

• Recommended: Use go2rtc with hardware acceleration
• Native mode: Reduce main_width, main_height, and main_fps in config
• Check CPU/GPU usage
• Ensure webcam supports requested FPS
• Check GET /api/stats (and GET /api/recording/status for recording) for per-output frame/drop counts if a specific consumer seems to be lagging

ONVIF discovery not working

• Check firewall allows UDP port 3702
• Verify local network allows multicast
• Use ONVIF Device Manager to test
• If authentication is enabled, make sure the ONVIF client is configured with WS-Security UsernameToken credentials matching username/password

License

MIT License - see LICENSE file for details

Credits

• Built with Python, NumPy, and OpenCV
• Uses go2rtc for RTSP streaming
• ONVIF protocol implementation based on WS-Discovery specs