Revision history [back]

Yes, I spent some days making it work. The current state of that code is not perfect :) The templates are available, you only have to insert some code to publish them.

If you only need detection and you have a fast CPU then this package might fit your needs after a bit of coding.

What this package can give you:

• fast SIFT-based feature matching for templates (fast: compared to other implementation)
• single detection results ("publishes" matched template)
• detection result names (now it's only printed I think but could get the file name of the template as a result)
• high false detection rate (based on my experiences)
• no pose estimation (important if you are considering grasping)

If you are further interested in this topic we can share experiences in private and create a summary for example on the ros wiki.