3D Machine Vision


3D reconstruction: the 3D shape of an object is reconstructed from the images of a two camera setup.

 

3D alignment: using a camera image and various image processing functions, the 3D pose of an object is determined. This data is used to control the robot.

3D object processing: Errors are detected using 3D surface inspection.

In industrial applications, e.g., within the Industrial Internet of Things, 3D vision is the acquisition, processing, and utilization of image-derived three dimensional data to affect or control a mechanical system or process. This allows to realize applications that could not be solved with 2D approaches. Technologies like “3D scene flow” can even be used to predict the motion of objects. As “eye of the production” MVTec’s software supports all aspects of 3D vision, such as:

  • 3D reconstruction
    Image acquisition Interfaces are provided for numerous commercially available 3D sensors.
    MVTec software products also offer tools for constructing 3D depth maps or point clouds from image techniques like multi-view stereo or sheet of light imaging. Furthermore the software  enables smoothing, sub-sampling, and triangulation of points for efficient processing and improved visualization.
  • 3D registration
    MVTec software allows customers to generate a complete 360 degree representation of an object by aligning 3D point clouds from multiple 3D images.

Video: 3D vision guided robotic assembly using surface-based 3D matching.

  • Surface-based 3D matching
    This tool searches for arbitrarily shaped 3D objects in a scene and determines their pose in 3D space. It locates multiple objects in a single scene even if objects are partially occluded or not entirely contained in the scene. This is a powerful tool for robotic bin picking and automated portioning systems as well as many more 3D machine vision applications.
  • 3D object processing
    Similar to blob analysis in 2D, MVTec’s software
    enables developers to measure and extract various features from 3D point clouds as well as segment point clouds based on those features. Background points can easily be removed via thresholding, and point clouds can be intersected by a plane to create a 2D cross-section profile.
  • 3D surface inspection
    Alignment of an acquired 3D point cloud object with a known object model or point cloud template enables users to check for and locate variations and deviations in the surfaces profile of manufactured parts.