Point Cloud Preprocessing
Before bin recognition, you need to preprocess the data by adjusting the parameters to improve the recognition accuracy and efficiency.
|
Except for the "Set recognition region," the parameters in the "Point cloud preprocessing" process are valid for all bins in the "3D Bin Recognition" tool. In other words, once the parameters are adjusted, the preprocessing results of all bins will change accordingly. If you want to use different point cloud preprocessing methods for different bins, you can add multiple "3D Bin Recognition" Steps to the project to set different point cloud preprocessing parameters for different bins to achieve different point cloud preprocessing results. |
Set Recognition Region (3D ROI)
-
Click Set and set a 3D ROI in the ROI setting window.
-
After the setting, click Save and apply.
Adjust Parameters
After setting the recognition region, you need to adjust the parameters related to point cloud preprocessing for processing the point cloud of the target object.
By default, only the Edge extraction effect parameter will be displayed. If you need to adjust more parameters, you can enable View more parameters.
|
Among the following parameters, Edge extraction effect and Noise removal level are valid for both edge and surface point clouds, while other parameters are only valid for surface point clouds. |
- Edge extraction effect
-
Description: Use this parameter to determine the edge extraction effect.
Value list: Fine, Standard, Rough, Extra rough, Custom
Instruction: Set this parameter according to the actual requirement. In general, select Standard. When Custom is selected, you need to set the normal variation threshold in degrees (°) for determining edge points. A point is considered an edge point if the normal variations of its neighboring points are above this threshold.
By normal variation means a comprehensive value considering the normal variation of the neighboring points.
- Noise removal level
-
Description: Use this parameter to determine the noise removal level.
Value list: None, Weak, Strong
Instruction: This parameter should be adjusted when there is severe noise or erroneous burrs on the edges of the point cloud of the target object. The stronger the removal level, the better the noise removal effect.
Point filter
- Min polar angle
-
Description: Use this parameter to determine the lower limit of the angular difference in degrees (°) in the process of point filtering. When the polar angle is lower than this value, the point will be filtered out.
Default value: 0°
- Max polar angle
-
Description: Use this parameter to determine the upper limit of the angular difference in degrees (°) in the process of point filtering. When the max polar angle is above this value, the point will be filtered out.
Default value: 70°
Remove noise by clustering
- Cluster radius
-
Default value: 3.000 mm
Description: Use this parameter to determine the radius in millimeters (mm) for clustering.
Instruction: When this parameter is large, points that are far apart will be grouped into the same cluster; when this value is small, points that are close together will be grouped into different clusters.
- Min point count per cluster
-
Description: This parameter is for filtering the results after clustering. Only the clusters with the number of points greater than the minimum point count can be output. When this parameter is large, the number of output clusters will decrease; when the parameter is small, the number of output clusters will increase.
Default value: 100
- Max point count per cluster
-
Description: This parameter is for filtering the results after clustering. Only the clusters with the number of points lower than the maximum point count can be output. When this parameter is large, the number of output clusters will increase; when the parameter is small, the number of output clusters will decrease.
Default value: 3000000
- Get highest layer clouds only
-
Description: This parameter is used to retain only the point cloud of the highest layer.
Default value: Disabled
- Layer Height
-
Description: This parameter is used to set the layer height of the highest-layer point clouds. Point clouds within this height will be retained.
Default value: 100.000 mm
Use "Segment target objects and bin" to Recognize Bins (Optional)
Based on the object-bin segmentation model package, this feature can segment target objects and bins from the input depth map and color image to generate target object masks and bin masks. This feature is suitable for identifying open-top turnover boxes of different colors and standard rectangular bins, but it is not suitable for identifying mesh bins or irregular bins.
If you want to use this feature, you can obtain an object-bin segmentation model package in the Download Center. If the recognition effect is unsatisfactory, you can import image data into Mech-DLK, label the images in Mech-DLK, and then train, validate, and export the model package.
After enabling this feature, you need to set the following parameters.
| Parameter | Description |
|---|---|
Model package file |
Parameter description: This parameter is used to select an imported object-bin segmentation model package.
|
ROI |
Parameter description: This parameter is used to set or modify the ROI.
|
Morphological transformation |
Parameter description: This parameter is used to perform morphological transformation on the inference result of the model package.
|
Kernel size |
Parameter description: This parameter is used to set the size of the operation kernel in morphological transformation, in pixels (px). The larger the kernel, the stronger the dilation or erosion effect.
|
View Preprocessing Result
After setting the above parameters, click Run Step or Run project to view the preprocessing result.
|
If you want to view other preprocessing results, you can switch relevant options above the visualization area. |
After point cloud preprocessing, click Next to enter the "Target object selection and recognition" process.