3D vision machine systems become mainstream

The completion of a recent project is evidence that the advent of the latest vision systems that have 3D capability offer substantial benefits to manufacturers.

The ability of these systems to build mathematical 3D models in real time means automation becomes simpler when processing the product on the production line.

There are several reasons why 3D machine vision has become mainstream.

First, the software-based machine vision systems that have the ability to transform 3D calibrations from 2D images are highly processor intensive.

This has been fed in turn by the availability of the multi-core processor.

Ten years ago, it could be argued that Microsoft Windows based control systems wouldn't have been taken seriously.

But this is the case today, and the fact that Windows XP takes care of image acquisition and simultaneous processing of multiple tasks or threads at OS level without the need for extra hardware means that the application software is left to get on with the job of 3D modelling.

One such application is the measurement of automotive bushes in 3D.

The host computer is a 3GHz quad-core industrial PC with Windows XP.

The system utilises seven cameras producing 19 images with LED lights and a laser dot-matrix light source.

The laser dot-matrix light is used to create points in the images for generating point clouds and the LED lights are used to create appropriate light conditions for other 2D and 3D measurements.

All cameras are individually calibrated to eliminate lens distortion and 3D calibrated using a known 3D object.

As a result, each camera is a true pinhole camera with a common 3D model.

The image processing is configured using standard 2D and 3D tools in Scorpion Vision software without any programming.

The vision configuration strategy was to measure individual 3D planes on the link arm containing the bushes using stereo vision and point cloud technology.

From the 3D planes the angle difference can be measured.

The planes are the basis for generating virtual 3D camera systems creating 2D image planes positioned for accurate 2D measurements on different parts of the link arm.

Stereo vision technology is used to create 3D coordinates of 2D points and then the 3D distance can be calculated.

A key feature in Scorpion Vision software is the easy way of creation and use of 3D models.

A 3D model requires a set of 3D points that can be used for finding 3D planes and geometrical structures like boxes and cylinders.

In this application, the dot-matrix laser light source generates bright dots that are found by using a blob tool in one of the images.

The output of the blob is a sequence of x and y coordinates of the dot centre points.

The order of the coordinates in the sequence is not significant.

The blob output is directly imported into the CreatePointCloud tool in Scorpion.

This tool has references to 3D calibration of all three cameras and can use either correlation or blob search technology to find and correlate the individual bright spots in all three cameras.

Finally by using the Scorpion stereo vision technology, the output of this tool is a sequence of x, y and z coordinates of the dots that can be plotted and displayed as an image.

The description belies the simplicity of the system.

The hardest part for the engineer is calibrating the cameras, which takes a little time but is a one off job.

Building the system is a matter of using the standard non-programming tools in the Scorpion tool set.

Once complete, Scorpion Vision then does all the work and outputs 3D numerical data to the associated control system.

Interfacing to Scorpion can be done using Ethernet RS232 serial bus or third-party modules.

Scorpion Vision has been integrated with most robot systems including Fanuc, Kuka, ABB, Denso, Bosch, Motoman, Kawaski and many more.

Scorpion Vision is running a seminar to promote 3D machine vision in food production at CenFRA in Doncaster on the 10 September that is free to attend.

Back to top