While machine vision is a very broad category, over the next few weeks we will post blogs relating to the segment that helps locate parts and guide robots (robotic guidance). Putting a finer point on it, in this post I'm writing about 3D perception for robotic applications.
There has been a good deal of discussion recently about 3D perception technology for robotic applications. Broad availability of 3D structured light sensors (the MicroSoft® Kinect® and now the KinectOne) at consumer prices (around $150) has enabled many researchers and developers to focus on 3D perception.
When considering if 3D vision is a good fit for an application, I always encourage folks to take another look at 2D vision. While the sensor itself might be relatively inexpensive, the software to solve specific applications with 3D vision can be expensive compared to traditional 2D machine vision systems. If 2D vision can get you where you need to go, why complicate things and drive costs up with 3D vision?
That said, there are times when 2D vision just won't do the job. For example, when part depth data is required.
While 3D machine vision systems have been applied to a broad range of applications covering several industries, there are a few basic factors that can help determine the approach including:
- Semi-structured vs. unstructured environments
- Known parts vs. unknown parts
- High-speed vs. low-speed
- Precise vs. imprecise placement
This following graphic illustrates the relationship between these factors and how they change from one application to another.
3D Vision in Manufacturing
Often, processes used in manufacturing applications such as machining, polishing and assembly, call for the part type and size to be defined and controlled. For example, in the bin picking process of a machining operation there is typically one part per production run; the part type and size is known (usually from a CAD model). The speed requirement is relatively low depending on the machine cycle time. While not a demanding high-speed application, the part must be placed precisely into the machine.
In assembly applications, where parts are well defined and controlled, both precision and high speed are often required.
3D Vision in Warehousing and Distribution Centers
At the other end of the spectrum are warehousing and distribution centers. Here, the size of cases removed from a pallet (depalletized) or unloaded from a truck may not be fully known. The 3D vision system must be able to anticipate a range of case sizes and respond accordingly.
Order picking in an e-commerce environment is another area where product type and size vary greatly. The system must handle a wide variety and, at times, an infinite number of parts. These applications require high speed, but are not at all precise.
Whether or not the part is moving (typically on a conveyor belt) or stationary is another important consideration for machine vision.
There are also trade-offs with buying a machine vision system designed for a specific application, or buying a tool set which provides more flexibility. More flexibility adds complexity and requires advanced understanding of vision theory and programming.
We'll discuss these and other machine vision topics in future blog posts.
By Tim DeRosett, Director of Strategic Initiatives