Vision systems for robot guidance
A vision system comprises a camera and microprocessor or computer, with associated software. This is a very wide definition that can be used to cover many different types of systems which aim to solve a large variety of different tasks. Vision systems can be implemented in virtually any industry for any purpose. It can be used for quality control to check dimensions, angles, colour or surface structure-or for the recognition of an object as used in VGR systems. A camera can be anything from a standard compact camera system with integrated vision processor to more complex laser sensors and high resolution high speed cameras. Combinations of several cameras to build up 3D images of an object are also available.Limitations of a vision system
There are always difficulties of integrated vision system to match the camera with the set expectations of the system, in most cases this is caused by lack of knowledge on behalf of the integrator or machine builder. Many vision systems can be applied successfully to virtually any production activity, as long as the user knows exactly how to set up system parameters. This set-up, however, requires a large amount of knowledge by the integrator and the number of possibilities can make the solution complex. Lighting in industrial environments can be another major downfall of many vision systems.VGR approaches
Typically, vision guidance systems fall into two categories; stationary camera mount, or robot arm-mounted camera. A stationary camera is typically mounted on a gantry or other structure where it can observe the entire robot cell area. This approach has the advantage of knowing its fixed position, providing a stable point of reference for all the activity within the cell. It has the disadvantage of additional infrastructure cost, and occasionally having its view obstructed by the robot arm's position. It also typically requires large image files (5 Mpixel or more) since the image must cover the entire work area. These may be 2D or 3D cameras, although the vast majority of installations (2019) are using machine vision 2D cameras offered by companies such as Keyence, Basler, Sick, Datalogic, COGNEX and many others. Emerging players such as Leopard Imaging, Pickit3D, Zivid, and Photoneo are offering 3D cameras for stationary use. COGNEX recently acquired EnShape to add 3D capabilities to its lineup as well. 3D stationary mount cameras create large image files and point clouds that require substantial computing resources to process. A camera mounted on a robot arm has some advantages and disadvantages. Some 3D cameras are simply too large to be practical when mounted on a robot, but Pickit 3D's Xbox cameras and 2D cameras such as Robotiq's wrist camera are compact and/or light enough to not meaningfully affect available robot working payload. An arm mounted camera has a smaller field of view, and can operate successfully at lower resolution, even VGA, because it is only surveying a fraction of the entire work cell at any point in time. This leads to faster image processing times. However, arm mounted cameras, whether 2D or 3D, typically suffer from XYZ disorientation because they are continually moving and have no way of knowing the robot arm's position. The typical workaround is to interrupt each robot cycle long enough for the camera to take another image and get reoriented. This is visible in essentially all published videos of arm-mounted camera's performance, whether 2D or 3D, and can increase cycle times by as much as double what would otherwise be required. Pickit 3D's Xbox camera has been arm-mounted for some applications. While capable of more complex 3D tasks such as bin picking, it still requires the stop-take-a-picture re-orientation mentioned above; it's 3D awareness does not help with that problem. Visual Robotics claims to eliminate this cycle interruption with their "Vision-in-Motion" capabilities. Their system combines a 2D imager with internal photogrammetry and software to perform 3D tasks at high speed, owing to the smaller image files. The company claims a pending patent covering techniques for ensuring the camera knows its location in 3D space without stopping to get reoriented, leading to substantially faster cycle times. While much faster than other 3D approaches, it is not likely to be able to handle the more complex 3D tasks a true stereo camera can. On the other hand, many 3D applications require relatively simple object identification easily supported by the technique. To date, their ability to visually pick objects in motion (e.g. items on a conveyor) using an arm-mounted camera appears to be unprecedented.VGR systems benefits
Traditional automation means serial production with large batch sizes and limited flexibility. Complete automation lines are usually built up around a single product or possibly a small family of similar products that can run in the same production line. If a component is changed or if a complete new product is introduced, this usually causes large changes in the automation process-in most cases new component fixtures are required with time-consuming setup procedures. If components are delivered to the process by traditional hoppers andSee also
* Machine vision * Simultaneous localization and mappingReferences
{{Reflist Industrial robotics Machine vision