The system function is to properly guide the robot destining it to the grasp of mechanical objects arranged randomly inside a caisson. This vision system, which arises as a solution to one of the most complex and most required problems in the industrial machine vision world, must locate objects easily accessible by the picking robot.
The innovative bin picking algorithm is the result of dedicated research years and development by Specialvideo within 3D. The three-dimensional data are “simplified” to obtain a model which can be applied to the classic and well-established 2D techniques such as pattern matching, quality control and dimensional measurements. The vision system images created by laser triangulation are then calibrated, that is, the pixels are converted into millimeters according to a reference system shared with the robot. In this way the coordinates exchange is correct and precise.
The bin picking algorithms also include selection criteria according to which the picking order is determined by the grasp and the exclusion of those elements whose grasp leads to a collision between the robot arm and with other parts or with the box.
3D objects localization
The three-dimensional objects reconstruction is realized through two cameras and a laser scanning. A laser projects a bright line on the scene framed by the cameras, so that the captured images contain the lines given by the intersection between the laser plane and the objects surfaces. Knowing the laser plane equation and the cameras position, through a triangulation process we obtain the information necessary to determine the points coordinates (x, y, z) belonging to the line projected on the object surface. This allows a substantial independence from the colors and surface object structure (texture). The images obtained are “dense”;
they contain three-dimensional information of the entire surface and allow to see even the small details in relief.
This technique allows a resolution and precision appropriate to the robot guide applications for large work areas.
Using two cameras allows to cover those that would be dead angles determined by the body edges and the objectsposition . There is also some flexibility in the body position relative to the robot itself.