is calibrated individually. ‘In a standard system,’ explains Mark Williamson, sales and marketing director at Stemmer Imaging, ‘if the camera is at a slight angle or the laser is not completely perpendicular it will affect the accuracy at different points in the 3D space. Metric 3D can achieve more than 10 times the accuracy of standard calibration methods, because the technique calibrates out any errors in the system caused by slight mechanical misalignment.’ CVB also includes Merge 3D, which uses data


from two cameras positioned on either side of the laser to give more information and avoid dead spots. Complex parts containing multiple peaks and troughs or sharp steps in their geometries can result in occlusions, in which the laser beam is blocked by the part. The two cameras are correlated and calibrated using Metric 3D, so that, even if each camera provides different area measurements, the software compensates for it, because every single point has been calibrated and cross-correlated.


‘Complex parts can result in occlusions in which the laser beam is blocked by the part’


Stemmer Imaging has recently implemented


a 3D system at a food processing application, in which the produce was imaged at 360° to get a very accurate measurement of the volume. ‘Many food products don’t have a fl at base and to get an accurate measurement of volume requires 360° 3D imaging,’ says Williamson. In this project, the Merge 3D capability was extended to use data from three cameras – one from the top, one from the bottom left, one from the bottom right. The food item moves over a conveyor with a gap in it and, using a complete laser sheet, both the top and underside could be imaged to create one integrated slice area and volume measurement. The system is calibrated to identify the alignment between the three cameras. ‘In theory, a system could be built without any errors if the cameras were positioning very exactly, but that’s impossible in a manufacturing environment,’ comments Williamson. ‘The calibration tool allows the system to not only be calibrated for area, but also to be calibrated for the interaction of multiple views of the same object.’


3D vision, 2D processing ‘Once the elevation map has been generated, a lot of traditional 2D tools can be applied to carry out the analysis,’ says Matrox Imaging’s Boriero.


An elevation map is effectively a 2D image in which the intensity information is replaced with the elevation information. A blob tool can be applied on a height map to fi nd the object, and multiplying the height with the area will give the volume, for instance. Murhed of Sick IVP comments: ‘Often, more specialised 3D tools are needed to engineer an effective application and a combination of 2D and 3D processing tools will typically give the best result.’ Robotics is one of the big areas for 3D


vision because of the fl exibility it provides to manufacturing. Random bin picking, for instance, is very hard to do with 2D cameras because objects cast shadows and are lying on top of each other at different angles. In the past, robots were blind workers, says Kreutzer of MVTec; they only operated at set coordinates. The integration of 3D vision means that the xyz coordinates of objects lying in random orientations are fed to the robot arm, allowing greater fl exibility for pick-and- place applications. One such system, using Point Grey’s Bumblebee XB3 stereo camera along with Cognex’s 3D-Locate software, was demonstrated at the Vision Show in Boston at the end of May. The system provided real-time 3D positional information to determine each part’s 3D orientation, including parts that were stacked or tilted, for such applications as vision- guided assembly, logistics, pick-and-place, and inspection. ‘Laser scanning is used heavily in industry,


while monocular/stereoscopic techniques are particularly suited to robotics applications,’ says Boriero. Stereoscopy uses two or more points of view to retrieve the 3D position of a feature. It requires that a given location be identifi ed in each view, which, according to Arnaud Lina, manager algorithm development team at Matrox Imaging, is the challenging aspect of the technique. Once the position is located, then, after calibration, lines of views can be projected in the 3D space and where these lines cross is the position of the part. One stereovision approach is dense


stereoscopy, where every pixel in the image is associated in 3D space. However, Lina explains that for a fl at surface it becomes impossible to associate one pixel in the fi rst view with another pixel in the second view correctly, resulting in areas with errors. A second approach operates by locating well-defi ned geometries in each view, such as edges or corners, and associating those between the views with a pre-reasoned knowledge of the part. ‘Not every pixel is associated, but only the pixels corresponding


IMVE Newsline


Receive the latest news and product information direct to your inbox every month


Sign up to IMVE Newsline, our monthly email newsletter


www.imveurope.com/ subscribe


Page 1  |  Page 2  |  Page 3  |  Page 4  |  Page 5  |  Page 6  |  Page 7  |  Page 8  |  Page 9  |  Page 10  |  Page 11  |  Page 12  |  Page 13  |  Page 14  |  Page 15  |  Page 16  |  Page 17  |  Page 18  |  Page 19  |  Page 20  |  Page 21  |  Page 22  |  Page 23  |  Page 24  |  Page 25  |  Page 26  |  Page 27  |  Page 28  |  Page 29  |  Page 30  |  Page 31  |  Page 32