How to speed up camera acquisition

Hi @sethg, the only way to improve the acquisition time is to reduce the processing load on the Robotiq Wrist Camera Vision System. Basically, the complexity of your background and the presence of many parts will increase the processing time for the Vision. We do not have any numbers to specs the processing time yet, we are evaluating this, it depends on a lot of factor.

Here are a few tips for a faster processing time :

Have a simple, uniform background. Presence of geometric shapes in your background is to be avoided
Have a stable ambient light. Although the system is quite robust to changes in the ambient lights, unstable conditions will increase the processing time.
Have less parts present in a snapshot. When the system takes a photo, it will process the image and identify the object with the highest success rate (defined by your detection threshold), if there is any way for you to reduce the number of parts present, it will greatly increase cycle time.

As an example, a simple case is a uniform clean background with a single object present, it takes roughly 300 ms to process. With a clustered field of view and a complex background (aluminium extrusion) I found out a process time of roughly 2-3 seconds.

One last thing, quite important:

Check your detection threshold, if it is set too high, the system won't match the object model to the snapshot you take. Thus you think it's slow but in fact it just does not recognize the part.

I usually target the minimum detection % on my object, and remove 5% from that to set it as the threshold. Also consider avoiding false positive, take a snapshot with no object present and make sure the system does not match the object. For example, if your object is match between 80 and 90% when you test, put a threshold of 75% and make sure that the object is not found when nothing is present.