tcxAzureKinect

Azure Kinect DK support for TrussC, implementing the tcxDepthCamera interface. Drive an Azure Kinect (a time-of-flight depth camera) through the same DepthCamera API as any other depth sensor.

Status: working on Windows (hardware-verified). Builds and links with the Azure Kinect Sensor SDK v1.4.1 (MSVC x64); example-basic runs against a connected device and renders a live colored point cloud — depth, color, and IR are all confirmed end-to-end. The example also shows color / depth / IR previews and orbits around the camera origin. Linux build wiring is present but not yet exercised. k4a has no macOS support. See example-basic/.

Requirements

• Azure Kinect Sensor SDK (k4a) installed. Officially supported on Linux and Windows only — not macOS.
  • Windows: install the SDK MSI from Microsoft. It installs to C:\Program Files\Azure Kinect SDK vX.Y.Z but does not ship a CMake config, so this addon's CMakeLists.txt locates the headers + k4a.lib under that path automatically (override with -DK4A_SDK_ROOT="…/sdk" if installed elsewhere). The runtime DLLs (k4a.dll + depthengine_*.dll) are copied next to the app executable at build time — no manual PATH setup needed.
  • Linux: install libk4a<ver>-dev (provides a k4a CMake package / libk4a), which is found via find_package(k4a).
• The tcxDepthCamera addon (this addon depends on it).

Usage

#include <tcxAzureKinect.h>
using namespace tcx;

shared_ptr<DepthCamera> cam = make_shared<AzureKinect>();  // device index 0
cam->setThreaded(true);     // grab on a background thread (call before setup)
cam->setup();
// ...
cam->update();
if (cam->isFrameNew()) {
    Mesh cloud = cam->toMesh({.colors = true});
    cloud.draw();
}

Color and IR are part of the canonical DepthFrame, so they are read directly on the camera (no capability cast needed):

if (cam->hasColor())    { const Pixels& c  = cam->getColorPixels(); }
if (cam->hasInfrared()) { const Pixels& ir = cam->getInfraredPixels(); }
const DepthFrame& f = cam->currentFrame();   // depth / world / color / ir / intrinsics

Notes

• This backend just fills the canonical DepthFrame in captureInto(); the tcxDepthCamera base provides all the accessors, meshing and threading.
• Units: depth distance and world coordinates are in meters (the tcxDepthCamera convention). Depth is stored as uint16 mm with depthScale = 0.001.
• Color is kept at its NATIVE full resolution (e.g. 720p) - it is NOT registered/downsampled to the depth geometry. The base computes the depth-> color mapping on demand from the color intrinsics + depth->color extrinsic (both cached from the k4a calibration), so getColorPixels() returns the full color frame and getColorTexCoordAt() / getColorAt() project correctly. Use registerColorToDepth() if you specifically want a depth-aligned image.
• Point cloud uses the SDK's depth_image_to_point_cloud transformation (accurate, accounts for lens distortion): the result is written to frame.world, and the base returns it from getWorldCoordinateAt() / toMesh() instead of re-deprojecting from intrinsics.
• IR is the active-brightness image, stored as a 1-channel F32 Pixels (read via getDataF32()).
• Sensor type reports DepthSensorType::ToF.

Configuration

The current first pass hardcodes NFOV unbinned depth, 720p BGRA color, 30 fps. Making these configurable is a planned follow-up.

License

MIT. See LICENSES.md. Depends on the k4a SDK (MIT, Microsoft) which must be installed separately.