3D Localization Configurator

3D Localization Configurator is a standalone application used to generate configuration file for Photoneo 3D Localization SDK. The configuration file contains the CAD of the searched object, defined gripping point and the localization parameters.

Localization Configurator requires PhoXi Control Application to connect to 3D scanner or file camera.

Overview

To download Localization SDK, visit our product webpage: Localization SDK

The Localization Configurator is installed together with the SDK. It serves as a simple wizard to generate PLCF file, which is used as an input for Localization SDK. The example code of how to use the SDK is located in installation folder, which is usually

• C:\Program Files\PhotoneoSDK\Localization\Examples\ in Windows
• \opt\PhotoneoSDK\Localization\Examples\ in Ubuntu

Copy the example directory outside the Program Files directory (because of the user rights) and use CMake to generate the project.

Setting the gripping point

The gripping point defines the point where the object should be picked by the gripper. Every gripper has a defined tool point. Gripping point and tool point together define the mutual position of the gripper and gripped object for bin picking applications.

In Localization Configurator, the gripping point has a position and orientation with respect to CAD model coordinate system. When the localization process finds the object, it returns the position and orientation of this gripping point (usually in robot coordinate space).

Localization parameters

Overlap with model

This percentage express how many points have to be seen on the object to consider the result correct

Higher percentage leads to higher accuracy and longer computation time

Amount of time

If checked, the localization process stops after the defined time limit

This constrain can be combined with "Number of instances"

Number of instances

If checked, the localization process stops after the defined number of objects is found

This constrain can be combined with "Amount of time"

Use Smart Memory

If checked, the localization process remembers the positions of found objects from the last run and uses this information in new search

This setting can strikingly speed up the localization process

The assumption for using this setting is that the scene is mostly static (only few parts change their position between consecutive searches)

Check occlusions only in a distance to Gripping Point

If checked, the calculation of the object occlusion is limited only to the area in the defined distance to the gripping point

By default (this option is not checked), object is considered occluded when there is another object on top of it (from the camera perspective)

The assumption for using this setting is that the object can be safely picked also when there are other objects on top of it (e.g. these objects will fall down without damage when the underlying object is being picked)

Advanced parameters

First the algorithm splits the scene into segments. Segment should be a continuous meaningful area in the scan. In the visualiser each segment has different shade of green color. When there are multiple objects in the scene, each object should have different shade of green. It does not matter when the object is split into several segments.

There are 3 settings affecting the segmentation: Scene clustering level - higher means more segments will be created Scene minimal cluster size - nominal number of points in segment. Smaller segments are filtered out (gray color) Scene maximal cluster size - nominal number of points in segment. Bigger segments are filtered out (gray color) Then there is an matching algorithm itself and another 3 settings: Matching algorithm: use either Surfaces (default) or Combined. Combined is good when the object is somewhat planar and have clear edges (it is not rounded). In your case Surfaces should work as well, but Combined might be slightly faster. Model Keypoints Sampling: the density of model subsampling. Higher density means more points and more extensive computations. For Combined matching algorithm the Dense option is not so heavy on the computation and might bring better results. Local Search Radius: the algorithm works on searching around numerous key points in the scene. This parameter sets the radius for searching. For specific situations, it might speed up or slow down the search, but should not have very significant effect if the part is found or not. The next 3 settings influence the tolerance of the matching to noise and accuracy of the fit: Feature fit consideration level: this is related to the segmentation. This number expresses the minimal size of the segment (as the percentage of object surface from specific viewpoint) to be this segment considered relevant for matching it with the object (Smaller segments are not relevant for fitting the object there). Maximal feature fit overflow: percentage of the segment which might not fit the object "perfectly" (the percentage is calculated from the size of the segment) Global maximal feature fit overflow: percentage of all segments which might not fit the object perfectly (the percentage is calculated from the sum of segment sizes) Fine alignment stage has also 3 settings: Fine Alignment Iterations: number of iterations of fine alignment algorithm. Meaningful numbers: 6 - 30 Fine Alignment Point Set: either Surface or Edge points. Edge points might be more useful for planar and rectangular objects, but usually remains on the Surface option. Fine Alignment Point Set Sampling: keep always Sampled, use Complete only in the situation when the point set is set to Edges The rest two options are for special cases: Projection Tolerance and Projection Hidden Part Tolerance - percentages, usually set to 100 %, which means it is turned off. Imagine you have a thin L shaped object and you scan it directly from the top, so you can see one side almost completely (except the shadows), but from the other side you see only the edge. In this case the localization might find the object flipped because it relies on the side that is visible and few points on the edge are not significant. By setting these tolerance to lower percentage, e.g. 10 - 45 % the localization will consider shadowing of the object on itself and further indicators to find correct "flip" of the part. The last option Overlap is taken from the main application window and the field Overlap with model: After the algorithm finds the object in the scene, it calculates number of points fitting on the CAD model. The model is "covered" with points only to some extent and this coverage we call overlap. When the overlap is lower than specified number, the object is not returned by the algorithm.

3D visualiser

3D visualiser helps to test the localization parameters on specific object. It displays the 3D point cloud acquired from 3D scanner or file camera via PhoXi Control Application.

After starting the localization process, the CAD models of found object are drawn into the point cloud. The color of object indicates whether the object is occluded (blue color) or not occluded (gold color). Gold objects are the most suitable to pick.

Controls

Left mouse button drag

rotate the scene

Mouse wheel or vertical right mouse button drag

Zoom in/out

Mouse wheel drag or SHIFT + left mouse button drag

move the scene

CTRL + left mouse drag

tilt the scene

Move mouse over some point and press F

Set the focus on this point

Press +

Increase the size of points

Press -

Decrease the size of points