Vision coordinate system

JGraafmans

Hi all,

I've been trying to develop a robust application using vision guided motion.
We have the following camera available: http://www.theimagingsource.com/en_US/products/zoom-cameras/gige-cmos-color/dfkz12gp031/

A quick sketch of the situation:
- A vacuum gripper is attached to the robot (ABB IRB 4600) to pick randomly placed bodies
- Those bodies are moved underneath a glue gun, which applies the glue to the body.
- The body is placed in a randomly placed housing.

The goal is to detect the bodies and housings on the carrier using a robot mounted camera.
Therefore we use the following approach:
- Take a picture of the complete carrier and send the position of all bodies and housings to the robot using socket messaging.
- Take a close-up picture of the body by moving the robot lower and to the correct x and y position.

The problem is now to get the correct x and y position from the second vision step.
In the first step we can use the outline of the carrier as reference position, however in the second step, the reference position is not visible anymore.

How are you handling those situations?

Thanks in advance,
Jeroen