Mech-Vision
Mech-Vision
Mech-Vision is the latest generation of Mech-Mind Robotics' innovative software for machine vision of three-dimensional point clouds. It is operated completely code-free via a graphical user interface.
Using intelligent algorithms such as 3D vision and deep learning, Mech-Vision performs complex tasks in a wide range of logistics and industrial manufacturing applications: From machine tending to bin picking in order picking and palletizing to automated gluing and joining of unsorted and even moving objects. With precise localization, reliable error detection and exact size measurement.
With Mech-Vision, you can process the 3D point clouds from 3D camera sensors and recognize objects using clustering, model and CAD matching, and deep learning. This provides you with information about the component location, component size and component properties. Based on this information, the best gripping points are then determined and transferred to the Mech-Viz robot controller.
Function
Mech-Vision allows easy processing of 3D point clouds without programming knowledge. Predefined functions can be simply dragged and dropped into the workspace to solve object recognition tasks. Pre-processing and optimization of point clouds can also be implemented in Mech-Vision in just a few editing steps.
Surface Segmentation identifies contiguous surfaces in the point cloud. Assuming that these points all belong to the same object, the center of the surface is output as the gripping point. This method is particularly suitable for applications with a focus on high speed and low accuracy requirements.
The Mech-Eye 3D camera generates a point cloud model of a component. An ideal gripping point is then defined on the object. This model is then searched for in the point cloud during operation. The method is accurate but requires more computing time than Surface Segmentation.
Instead of a sensor model, CAD matching uses an available CAD model. The advantage: compared to camera-generated models, such computer-generated templates are fully defined and usually very detailed. The imaging is free of camera noise, which shortens computing times.
Deep Learning uses 2D data to train a neural network that identifies exact positions in the point cloud. Different methods such as classification, instance segmentation or grasp point prediction are used.
Mech-Vison includes an integrated editor that can be used to automatically edit point clouds.
Many applications require the detection of the load carrier (container) before a component can be picked out of it. Mech-Vision, in conjunction with Mech-Viz, determines the exact position of the load carrier as required. The collision model is thus automatically adapted to the current position of the load carrier.
Artificial intelligence
Let the camera decide: Mech-Eye precisely delivers all the necessary data for classification, segmentation or gripping point prediction where classic machine vision algorithms reach their limits. Use pre-trained neural networks for this or simply train them yourself with the Mech-Mind Deep Learning Kit. Completely locally in your system environment - without access to the cloud.