3D Imaging – Locating Multiple Objects in a PointCloud

One of the most challenging tasks in machine vision has been “Random Bin Picking” – RBP. At Tordivel we do not think there is one single solution for this task. Depending on the part shape and more – MonoPose3D, Stereo Vision or 3D imaging can be a solution. In 2006 we started a large R&D project Auto3D together with Kongsberg Automotive, SINTEF and Conoptica to develope a set of tools to conquer RBP. We are now at the end of the R&D project and are proud of our achievements. We have moved Scorpion Vision Software from version 5.0 to version 8.0 and turned the product from a robust 2D package to a complete cutting-edge 3D framework including:

  • 3D visualisation
  • A 3D toolbox containing 50 tools
  • Complete support for 3D stereo vision – including numerous 2D improvements to support stereo vision
  • A nice tool set for point cloud processing including ConnectedComponents, PlaneFit, SphereFit and CylinderFit.
  • 3DMaMa an extremely powerful tool to multiple objects in a 3D point cloud
  • An GCPS – Gray Code Phase Shift Software Module to create 3D models using stripelight and a standard projector

Read more :  ….

Random Bin Picking Scene

The image below shows a typical automotive part to be picked from a bin.

Stabel_2_shutter600

3D Model captured using Scorpion GCPS

The model shown is captured using Scorpion GCPS. The image capture took 7.5 seconds. The model contains 1.2 million 3D points spanning an xyz-area of 1000 mm x 700 mm x 300 mm. The height resolution is 0.1 mm.

objects

Object Location using 3DMaMa

On a duo core Intel CPU 3DMaMa has located 5 parts in 3D, xyz and three angles,  in 1.8 seconds. On a Quad-Core processor it will be located in less than 1.0 second.

3dobject_located

The basic principle is to select the center of two planes on the parts. 3DMaMa then runs an exhaustive search to locate these plane within the 3D image. For each candidate it calculates and quality measure and match percentage. The dots in the image maps the model on top of the image to show where the parts a located.

How to teach 3DMaMa?

3DMaMa locates the parts based on a 3D Model. The model can be imported from a true 3D Model or be created based on points capture by Scorpion GCPS.

3dmama_template

Conclusion

3DMaMa is available to beta tester of Scorpion 8.0. We at Tordivel believe this is a kind of breakthrough in 3D part location.

Scorpion GCPS, MonoPose3D and 3DMaMa are trademarks of Tordivel AS.

3 Responses

  1. Is it possible to use point clouds created by other means than GCPS? Like different kinds of laser based scanning systems or TOF cameras?

    • The pointcloud format is open. Any 3D device can generate a pointcloud to be processed by Scorpion, but note that some tools required an ordered point cloud. This means that x,y are ordered. With GCPS there is one 3D coordinate pr pixel, thus a sxga camera will generate 1280 x 1024 approx 1.2 Mbyte 3D coordinates.

  2. […] regular contributions by Thor Vollset, the man behind Scorpion Vision. Recent entries of note are 3D Imaging – Locating Multiple Objects in a PointCloud, which discusses the ability of the software to identify and differentiate multiple objects […]

Leave a Reply

Please log in using one of these methods to post your comment:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: