Skip to content

Target - CAD

Given the CAD target and the RGB image of the scene, the tracker estimates the 6 degree-of-freedom (DoF) pose of the target to the camera. The estimated pose can be used to superimpose a rendered version of the target onto the image and add details or other augmented reality effects.

Finding the CAD target in the image requires an initial detection phase, that computes a rough estimate of its 6 DoF pose. Once the detection is successful, the tracker refines and updates the pose in real-time for every frame. If the tracking fails or the tracked target goes out of view, the tracker waits for the re-initialized detector to return a rough pose estimate of the target.

Currently, CAD targets cannot be trained with the Trainer CLI 1.8.0, please refer to 1.7.0 CLI trainer or download available sample targets

Track supports up to 3 different CAD targets:

Limitations

Tracking performance depends on many variables. Achieve the best performance by following the guidelines in the following sections.

CAD-Targets should

  • be rigid (non-deformable)
  • have less than 20.000 faces
  • avoid self-similarity

The CAD-Target should

  • not be occluded by the environment
  • not cast a hard shadow onto the environment

For best tracking performance, follow the following recommendations regarding the relationship of the target and the camera

  • Training and tracking use the same camera calibration parameters
  • The camera/target moves smoothly and slowly compared to the camera's frame rate (avoid abrupt movements)
  • Only one target is visible at a time
  • Detection is limited to the defined training parameters
  • The object is well visible

Multi-CAD Limitations

Multi-CAD performance depends on the complexity of each CAD-Target, it should be avoided to use targets with a cumulative number of faces above 20.000.