A playground to understand how the positional tracking systems for PlayStation VR, Oculus Rift, and HTC Vive work.
C Meson C++
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
src
tools
xml
.gitignore
AUTHORS
COPYING
COPYING.LIB
README.rst
meson.build
meson_options.txt

README.rst

ouvrt

  1. About
  2. Setup and build
  3. ouvrtd
  4. Tools
  5. Todo

1. About

ouvrt is a playground to understand how the positional tracking systems used by the PlayStation VR, Oculus Rift (DK2, CV1), and HTC Vive virtual reality headsets work. The main component is the ouvrtd daemon that detects and opens relevant USB devices and sets them up for tracking.

Currently the following devices can be detected:

  • Lenovo Explorer headset (USB)
  • PlayStation VR headset (USB, via PSVR processing box)
  • Rift CV1 headset (USB)
  • Rift DK2 headset (USB)
  • Rift DK2 Positional Tracker (USB)
  • Rift remote (wireless via Rift CV1 headset)
  • Rift touch controller (wireless via Rift CV1 headset)
  • Vive headset (USB)
  • Vive base station (optical via Vive headset or controller)
  • Vive controller (USB or wireless via Vive headset)

Features are still limited to enabling the tracking LEDs for PlayStation VR, Rift DK2 and CV1, setting up the camera sensor for synchronized exposure (DK2 only), and capturing video frames into a GStreamer pipeline for debugging (DK2). IMU sensor data is captured and can be sent along with axis and button state via UDP.

2. Setup and build

The following prerequisite libraries and development packages are necessary to build ouvrt:

  • GLib/GObject/GIO
  • GStreamer (optional)
  • JSON-GLib
  • OpenCV (optional)
  • libudev
  • Meson
  • zlib

On a Debian stretch system these can be installed with the following commands:

$ apt-get install build-essential libglib2.0-dev libjson-glib-dev \
  libudev-dev meson pkg-config
$ apt-get install libgstreamer-1.0-dev libopencv-dev

To configure the build system and build everything, follow the standard Meson build procedure:

$ meson builddir
$ cd builddir
$ ninja

To build without GStreamer or OpenCV dependency, disable the corresponding options before calling ninja:

$ cd builddir
$ meson configure -D gstreamer=false -D opencv=false

3. ouvrtd

Make sure you have permissions to access the /dev/hidraw and /dev/video devices corresponding to the Rift DK2 and the DK2 Positional Tracker. Then run ouvrtd:

$ ./ouvrtd

If compiled with GStreamer support, the daemon will create a shared memory socket /tmp/ouvrtd-gst-0 and, if a DK2 Positional Tracker is connected, write frames into it as soon as a GStreamer shmsrc connects to it. To see the captured frames, run:

$ gst-launch-1.0 shmsrc socket-path=/tmp/ouvrtd-gst-0 is-live=true ! \
  video/x-raw,format=GRAY8,width=752,height=480,framerate=60/1 ! \
  videoconvert ! autovideosink

4. Tools

The dump-eeprom tool reads the Positional Tracker DK2 EEPROM and writes it to a file or stdout:

$ ./dump-eeprom camera-dk2-rom.bin

$ ./dump-eeprom - | hexdump -C

5. Todo

  • Add blob detection and tracking
  • Enable Rift DK2 IR LED blinking patterns
  • Add individual blinking LED detection to the blob tracker
  • Add a 3D model of the tracking LEDs, readout from the Rift
  • Add support for camera intrinsic and lens distortion parameters, readout from the Rift and/or camera EEPROM
  • Add a PnP solver to estimate the pose from 3D-2D point correspondences
  • Add RANSAC PnP solver support for the initial pose estimation
  • Feed the projection of the estimated pose back as starting points for the PnP solver and blob tracker
  • Add Rift DK2 IMU support to estimate the pose from integrated gyro and acceleration sensor readouts
  • Add sensor fusion, correcting the IMU pose from the camera pose regularly, use the fused pose estimate to feed back into PnP solver and blob tracker
  • Implement proper time handling for all of this