D435 parcel dimension measurement: How can I reduce errors caused by reflective packing tape?
We are using a RealSense D435 to measure the length, width, and height of parcels at a warehouse receiving station. Our target is an absolute error of no more than 10 mm for each dimension.
Our setup:
- The camera is mounted above the measurement table, pointing vertically downward.
- Parcels are cardboard boxes with packing tape, including shiny and wrinkled areas.
- The workspace has overhead ceiling lights.
Reflections from the tape appear to be affecting measurement accuracy. Reducing the lighting also seems to increase errors, so we would like to find a suitable combination of lighting, camera settings, and optical filters.
Could you advise on the following?
- How can we determine whether the errors are caused by ambient-light glare, reflections from the IR projector, or another stereo-matching issue? What should we look for in the left/right infrared images and depth map?
- Would polarizing filters on both stereo cameras help with this top-down setup? I have purchased a Kase 46 mm photographic CPL filter, but its performance around 850 nm is not specified. Is this type worth testing, or would you recommend a particular linear polarizer or IR-pass filter?
- Which lighting arrangement would you recommend for taped parcels: low-angle side lighting, diffuse lighting with crossed polarizers on the lights and cameras, or another approach? Would coaxial lighting be suitable?
- Which exposure, gain, projector power, or depth settings should we check first? Are there recommended processing methods for rejecting reflection-related outliers without distorting the parcel dimensions?
Is our 5 mm accuracy target realistic with the D435 for this application, and what working-distance and parcel-size limits should we consider?
Please let me know which additional details or sample captures would be most useful for troubleshooting.
Thank you for your help!
-
Hi LEWIS LEWIS Thanks very much for your questions.
- A key indicator of areas of the image where the camera had difficulty reading a surface due to glare is if there are empty black areas (on the depth image) or areas of bright white (on the IR image). The IR sensors can see in the dark, so if you turn the overhead lighting off and there is still glare on the IR image, this could indicate that the IR light projection is responsible.
- Yes, placing thin-film linear polarizer filters over the camera lenses on the outside of the camera can greatly negate the glare from reflections. Most polarizing filters will work as long as they are linear, though circular filters like those in 3D glasses do not work. The filters can be purchased inexpensively on stores such as Amazon by searching for ‘linear polarizing filter sheet’.
- LED lights would be recommendable as they do not cast shadow. Shadows can cause problems with measuring cardboard boxes from overhead because the floor shadow confuses the camera into thinking it is looking at a box edge.
- The default settings should work well, though you can experiment with reducing the Laser Power, as this will decrease the intensity of the IR light projection and so may reduce reflections from the tape.
Although decreasing Laser Power can reduce the quality of the depth image, having a good level of artificial overhead lighting can help to compensate for this. The default value of Laser Power is 156, so you could try setting it to ‘99’.
It is possible to automatically eliminate outliers with software filters, though this typically involves using a third-party library such as PCL in combination with the RealSense SDK as these functions are not built into the SDK.
Regarding the 5 mm accuracy target, if the camera is 1 meter or less above the box then it may be an attainable target for D435. If the camera is further from the box then accuracy is likely to be less.
There is a detailed discussion about measuring boxes from overhead at the link below.
https://github.com/realsenseai/librealsense/issues/14863
The customer in that case found that they experienced measuring inaccuracy if the box was larger than the weigh scale platform beneath it so that the edges of the platform were not visible to the camera. They found that a solution that worked for them was to capture the empty weigh platform first and then capture again when the box was placed on the platform.
Please sign in to leave a comment.
Comments
1 comment