libfreenect2  0.2
API Reference

Older versions: 0.1.


libfreenect2 is an open source cross-platform driver for Kinect for Windows v2 devices. For information on installation and troubleshooting, see the GitHub repository.

This documentation is designed for application developers who want to extract and use depth and color images from Kinect v2 for further processing. Additional questions and comments not covered by this documentation can be posted to GitHub issues.

This documentation may require some understanding on camera calibration and 3-D geometry.


Issues and Future Work

Getting Started

To read the API documentation, start with the Modules page which nicely organizes classes according to their functionalities.

Example programs can be found in the source distribution under the examples directory. There also includes an example CMake build system for a standalone application that uses libfreenect2 binary installation.

Many internal details are hidden from this public API. For details on Kinect v2's USB protocols, depth decoding algorithms, calibration algorithms, and how to implement performance optimizers, you are encouraged to read the source code. The source code is the updated and authoritative reference for any functionalities.

Environment Variables

There are a few environment variables providing controls for both end-users and programmers:

LIBFREENECT2_LOGGER_LEVEL: The default logging level if not explicitly set by the code. LIBFREENECT2_PIPELINE: The default pipeline if not explicitly set by the code. LIBFREENECT2_RGB_TRANSFER_SIZE, LIBFREENECT2_RGB_TRANSFERS, LIBFREENECT2_IR_PACKETS, LIBFREENECT2_IR_TRANSFERS: Tuning the USB buffer sizes. Use only if you know what you are doing.

You can also see the following walkthrough for the most basic usage.


Here is an example to walk you through the API. See examples/Protonect.cpp for the full source.


First, include necessary headers. registration.h and logger.h are optional if you don't use them.


This shows how to set up the logger and logging level.

Though libfreenect2 will have an initial global logger created with createConsoleLoggerWithDefaultLevel(). You do not have to explicitly call this if the default is already what you want.

You can implement a custom Logger and redirect libfreenect2's log messages to desired places.

Here is an example to save log messages to a file.

#include <fstream>
#include <cstdlib>
class MyFileLogger: public libfreenect2::Logger
std::ofstream logfile_;
MyFileLogger(const char *filename)
if (filename);
level_ = Debug;
bool good()
return logfile_.is_open() && logfile_.good();
virtual void log(Level level, const std::string &message)
logfile_ << "[" << libfreenect2::Logger::level2str(level) << "] " << message << std::endl;

And use it

MyFileLogger *filelogger = new MyFileLogger(getenv("LOGFILE"));
if (filelogger->good())
delete filelogger;

libfreenect2 uses a single global logger regardless of number of contexts and devices. You may have to implement thread safety measure in log(), which is called from multiple threads. Console loggers are thread safe because std::cout and std::cerr are thread safe.

Initialize and Discover Devices

You need these structures for all operations. Here it uses only one device.

You must enumerate all Kinect v2 devices before doing anything else related to devices.

if(freenect2.enumerateDevices() == 0)
std::cout << "no device connected!" << std::endl;
return -1;
if (serial == "")
serial = freenect2.getDefaultDeviceSerialNumber();

Also, you can create a specific PacketPipeline instead using the default one for opening the device. Alternatives include OpenGLPacketPipeline, OpenCLPacketPipeline, etc.

Open and Configure the Device

Now you can open the device by its serial number, and using the specific pipeline.

dev = freenect2.openDevice(serial, pipeline);

You can also open the device without providing a pipeline, then a default is used. There are a few alternative ways to openDevice().

After opening, you need to attach Framelisteners to the device to receive images frames.

This SyncMultiFrameListener will wait until all specified types of frames are received once. Like loggers, you may also implement your own frame listeners using the same interface.

int types = 0;
if (enable_rgb)
if (enable_depth)

You cannot configure the device after starting it.

Start the Device

After finishing configuring the device, you can start the device. You must start the device before querying any information of the device.

if (enable_rgb && enable_depth)
if (!dev->start())
return -1;
if (!dev->startStreams(enable_rgb, enable_depth))
return -1;
std::cout << "device serial: " << dev->getSerialNumber() << std::endl;
std::cout << "device firmware: " << dev->getFirmwareVersion() << std::endl;

You can setIrCameraParams() after start if you have your own depth calibration parameters.

Otherwise you can also use the factory preset parameters for Registration. You can also provide your own depth calibration parameterss (though not color camera calibration parameters right now). Registration is optional.

libfreenect2::Frame undistorted(512, 424, 4), registered(512, 424, 4);

At this time, the processing has begun, and the data flows through the pipeline towards your frame listeners.

Receive Image Frames

This example uses a loop to receive image frames.

while(!protonect_shutdown && (framemax == (size_t)-1 || framecount < framemax))
if (!listener.waitForNewFrame(frames, 10*1000)) // 10 sconds
std::cout << "timeout!" << std::endl;
return -1;

waitForNewFrame() here will block until required frames are all received, and then you can extract Frame according to the type.

See libfreenect2::Frame for details about pixel format, dimensions, and metadata.

You can do your own things using the frame data. You can feed it to OpenCV, PCL, etc. Here, you can perform registration:

registration->apply(rgb, depth, &undistorted, &registered);

After you are done with this frame, you must release it.


Stop the Device

If you are finished and no longer need to receive more frames, you can stop the device and exit.


Pause the Device

You can also temporarily pause the device with stop() and start().

if (protonect_paused)
protonect_paused = !protonect_paused;

Doing this during waitForNewFrame() should be thread safe, and tests also show well. But a guarantee of thread safety has not been checked yet.