The visual detection and recognition of sought-after workpieces or components is a central challenge for the automation of many processes. Only if robot systems reliably and precisely determine the position and presence of parts in bulk or ordered layers, on pallets or conveyor belts, can they react to the situation and handle parts flexibly. A key technology for this is industrial image processing.
Powerful 3D cameras such as Ensenso from IDS and tools that efficiently process the generated data, such as matching algorithms, are now making practical implementations of robot bin picking tangible.
Typical challenges for camera-based methods to recognise complex geometries or hard-to-distinguish parts in unsorted containers and multi-layer arrangements are poor light and contrast conditions. In addition, objects with reflective or specular surfaces often do not produce enough texture detail to provide 3D images of sufficient quality for further processing. The generation of high-quality 3D data is demanding and the required resolution is reflected in the budget for the necessary components. Here, 2D camera technology provides a basis for high resolutions as well as suitable acquisition speeds. The runtime of the entire gripping process is relevant for the possible achievable cycle times of the robot, which are often considered a measure of efficient automation
Skillfully balancing efforts, costs and quality
The actual processing of the 3D data into 6-dimensional poses is also technically complex and the algorithms involved generally have to be adjusted with many parameters over a long period of time until they function satisfactorily. This is where it usually becomes apparent quite quickly how user-friendly and flexible a system can be. In order to be able to guarantee short (re)set-up times, the integration effort must remain simple and manageable even if the objects are changed or their localisation by the software tools. This is the only way that robot-supported automation can make the most of its adaptability, even for small batches or one-off production. In this context, it is also important to be able to realise an evaluation or proof-of-concept with little time and expense. But only by skilfully applying individual algorithms through the use of a lot of experience can software tools create a balanced relationship between effort, costs and quality without requiring the skills of specialists every time.
Finding CAD models in point clouds thanks to Ensenso
The new Ensenso PartFinder integrates different approaches from 2D and 3D-based methods from the outset in order to use all the advantages for finding solutions. 2D features provide textures, edges or shadows, which means that not only complex shapes, but also shapes with difficult 3D views, such as very flat parts or holes, are recognised quickly and stably. The template-based matching process is also capable of locating several of the sought-after parts in a single run, which is very important for the efficiency of the subsequent gripping processes and thus also the overall runtime.
Ensenso users have the flexibility to go from low cost for geometrically correct, relatively low-resolution scenes with the Ensenso S to extremely high accuracy even over large volumes and distances with the Ensenso X or XR series cameras. Whether a quick evaluation by means of an easy-to-use GUI or a complete integration into existing software structures is required, with the Ensenso PartFinder, even tricky localisation tasks can now be mastered robustly and easily through the combined use of depth information, surface standards and texture data.
Soon, IDS Imaging Development Systems will be launching a new 3D camera model in the Ensenso product line. With Ensenso C, the company offers a camera that for the first time delivers both 3D data and 2D colour information from one device.