Troubleshooting

Due to incorrect setting of 3D ROI, the carton(s) is beyond the 3D ROI. Hence, no result can be output from the Mech-Vision project.

In the Point Cloud Preprocessing & Get the Mask of the Highest Layer Procedure, the carton dimensions entered in the Step Parameters panel differ significantly from the recognized carton dimensions, making dimension validation impossible and resulting in failure to locate the carton.

Deep learning failed to correctly segment masks of individual cartons, making dimension validation impossible in the subsequent processing and resulting in failure to locate the carton.

Reset 3D ROI based on workstation layout and pallet pattern dimensions, so that the cartons will be within the ROI.

Measure the carton dimensions and compare to the entered dimensions. If the difference is huge, you can adjust the parameters under Read Object Dimensions in the Step Parameters panel of the Point Cloud Preprocessing & Get the Mask of the Highest Layer Procedure to reset the carton dimensions. If the difference is small, you can appropriately increase the value of the Box Dimension Error Tolerance parameter in the Calculate Carton Poses Procedure.

Double-click the Segment Masks of Individual Cartons Using Deep Learning Procedure, check the visualized output of the Deep Learning Model Package Inference Step, and ensure that the visualized output is normal, as shown in the figure below. If the visual output results are abnormal, please refer to Problem 3.

The ambient light on site was too strong.

The setting of camera 3D exposure parameter was not proper.

When the ambient light is too strong, it is recommended to use longer exposure time to enhance the image quality. When the ambient light is too strong and it is not effective to increase the exposure time, you can consider shading the station so that the ambient light intensity at the station remains constant.

As the captured image tends to be too bright when the carton is white, it is recommended to use shorter exposure time; as the captured image tends to be too dark when the carton is black, it is recommended to use longer exposure time.

When there are cartons with different colors, it is recommended to capture images using the method of combining shorter exposure time and longer exposure time.

When capturing 2D images, the camera exposure time and white balance settings were not proper. As shown below, the image on the left is captured when the camera exposure time is normal, and the image on the right is captured when that is abnormal.

The deep learning recognition region (2D ROI) was not set.

Masks of individual cartons obtained by using deep learning are abnormal. In this case, you need to iterate your deep learning model.

First, you need to confirm whether the 2D image has problems such as overexposure, underexposure, shadows, abnormal white balance, etc. If the above problems exist, you need to adjust the camera exposure time and white balance to ensure that the captured 2D image is normal. After ensuring that the 2D image is normal, if the recognition performance is still poor, please refer to the next part.

Reset 2D ROI.

Acquire carton image data to iterate your deep learning model.

Adjust the Z-axis of the carton poses to a specific orientation in the Mech-Vision project.

Set soft limits.

In the Vision Move Step, make sure to select the box of Remove Point Cloud of Target Workobject in the Remove Point Cloud of Target Workobject parameter group.

In the Vision Move Step, check whether the parameter values of XY-Plane Point Cloud Removal Range Expansion and Z-Direction Point Cloud Removal Range Expansion are too small in the Remove Point Cloud of Target Workobject parameter group. If the parameter values are too small, increase the parameter values appropriately.