3-D vision made simple
By Larry Maloney, Contributing Technical Editor -- Test & Measurement World, 3/1/2009 2:00:00 AM
|
|
|
Most machine-vision applications require only a 2-D solution, but 3-D applications have been increasing over the last five years as manufacturers add more automation to their production lines.
“One of the biggest drivers for 3-D is the move toward greater use of robotics,” said Heiko Eisele, president of MVTec, a Munich-based machine-vision software company with US offices in Cambridge, MA. “While companies need the increased throughput that robots deliver, they also need a human’s ability to recognize a greater variety of objects in three dimensions, and that means integrating a vision system with the robot application.”
This may sound like an expensive proposition, but it doesn’t have to be. Traditionally, such 3-D applications require two cameras, which use a triangulation technique to acquire 3-D image data. But with advancements in software algorithms and precise camera calibration, you can achieve 3-D vision with just one camera, noted Eisele. “All you need is a standard machine-vision camera, an illumination source, a host PC, and application software,” he said.
 In training a single-camera 3-D vision system, users combine a calibrated camera with an object model to determine a specified range of poses that a part will present to the camera. |
Take the example of an application that calls for a robot to pick a part from a bin and feed it to the next station on a production line. As Eisele explained, you must first determine which parts are suitable for grasping by the robot and then decide the appropriate grasping points. To accomplish that step, MVTec’s customers use the company’s Halcon application software to import a 3-D CAD file of the part, from which it creates an “object model” of the part.
Along with creating the object model, users also must calibrate the camera. Many calibration procedures only compute the size equivalent of a pixel or interpolate between known grid points. But that approach isn’t sufficient for 3-D vision, said Eisele. You must include the intrinsic camera parameters in the calibration, such as the focal length and distortion coefficients, as well as the position and orientation of the camera in a world coordinate system.
During the training process for the vision system, users combine the object model with the calibrated camera to determine a specified range of poses that the part will present to the camera in six degrees of freedom. “This is what we call the 3-D matching process,” explained Eisele. “You basically are creating one contour-based model for each possible camera pose, but the software does it automatically for you, including the camera calibration step. We’ve simplified the process so users can focus on the parameters of the application.”
MVTec is approaching robot OEMs and system integrators with this single-camera 3-D vision concept. Among the many possible applications: guiding robots for welding and sealing operations. Eisele pointed out, however, that the single-camera technique is not suitable for measurement tasks, such as determining the thickness of solder on a part. “This requires a two-camera solution,” explained Eisele, “because you are not working with a known object. You can’t measure and determine position at the same time with one camera.”
No related content found.
- 0 rated items found.
Datasheets.com Electronic Parts & Inventory Search
185 million searchable parts
- Part Number
- Description
- Inventory
- Products
- Manufacturers
View All Webcasts
