CANDELOR is using a very generic approach for object localization. This has the huge advantage that a large category of objects are possible to recognize — CANDELOR supports both highly complex objects and simple objects like this cylindrical object. The configuration is exactly the same as for the complex object:
- Take a CAD file of the object,
- use it to automatically generate configuration settings, and
- use this configuration for object recognition.
Optionally, configuration settings can be fine tuned for better performance.
In the above example, CANDELOR is used to detect a cylindrical object in a scan of about 150,000 3D points. As an added difficulty, one side of the object has an elevated border.
CANDELOR has correctly localized 72 parts in a few seconds on a standard desktop PC. To visualize the result quality, we visualize the distance between the detect object from the points with colors. Blue points represent an exact fit, yellow points are further away, and red dots are far away from any detected object. In this example the localization result is used in the robotic bin picking system IRobFeeder.