4. Troubleshooting
MiR AI Camera Getting started (en) 09/2019 - v.1.0 ©Copyright 2019: Mobile Industrial Robots A/S.
42
4. Troubleshooting
This section describes possible reasons why your MiR AI Camera may not be detecting
objects reliably. With each reason, there is a workaround or solution that you should apply to
avoid or solve the issue.
•
Not enough training data
During the Collection phase, if the target objects did not enter the camera's field of view,
you may not have collected enough images for training. If the Validation phase did not
include many images of the target objects, you should set the cameras to run the Col-
lection phase again and make sure that the objects enter the cameras' field of view more
frequently. It is recommended that there are at least 150 well framed images labeled for
each target object.
•
Change in lighting after collection phase
If the lighting has changed significantly since the Collection phase, the camera may not be
able to recognize the objects. Run the Collection phase again, while the lighting conditions
are the same as when the AI camera is operating.
If the lightning conditions around your MiR AI Camera change throughout
the day, make sure to run the Collection phase in all the different lighting
conditions.
•
Incorrect labeling
If you have labeled any images incorrectly, the data model may be corrupted. Repeat the
Validation phase or review all the labeled images in the training server to ensure that all
images have been labeled correctly.
•
Inclusion of too many invalid image samples
If invalid images are used in the Training phase, the data model may be corrupted.
Repeat the Validation phase or review all the labeled images in the training server to
ensure that the validated images follow the guidelines for valid sample images.
•
Objects have changed appearance
If the objects you have trained your MiR AI Camera to detect have changed appearance
significantly, MiR AI Camera may no longer be able to recognize them. Run the Collection
phase again and make sure the new objects are included.
