Optimizing Mech-Mind Vision System for Faster Picking in Mixed Pallet Setup

Knowledge Base
1

Hello all,

I am currently undertaking a project using Mech-Mind’s 3D vision system for robotic picking from a mixed pallet environment. The hardware is working well in my experience, but I am encountering some lag or time lost when running object detection. This is especially noticeable in mixed object pods, and with lighting changes and overlapping objects. I am using Mech-Eye LSR and Mech-Vision for my workflow.

My questions are: Is there a best practice - tuning the ROI or reworking the training data with increased variability to optimize for detection speed and reliability in such mixed object use case? Also, does a more powerful GPU greatly affect Mech-Vision processing time?

I came across this website:https://community.react-js-online-training-mech-mind.com/t/topic/356

I have read some documents in the Mech-Mind developer hub but I’m most interested in learning from people who have dealt with similar issues in real-life situations. Any practical tips or workflow changes that helped in your experience?

Thank you in advance - look forward to your opinions.
williamclark

2

Which Exposure Mode did you use in Mech-Eye Viewer? This appears might to be caused by Auto mode. The Auto mode results in longer time, because the camera need more time to calculate the exposure time, ensuring the image reaches the target gray value. Choosing Timed mode would be faster.

“tuning the ROI or reworking the training data with increased variability” can optimize for reliability but not detection speed. And “more powerful GPU” will improve Mech-Vision processing speed.

And if you want to impove Mech-Vision processing speed, maybe you can decrease the “input image size” when you rework the DL model with Mech-DLK.

image
image1920×995 199 KB