A Hybrid Approach for Object Localization Combining Mask R-CNN and Halcon in an Assembly Scenario

Robust and accurate object localization is a key component in the context of robotic grasping. Especially in collaborative assembly scenarios it is important that the robot is able to detect objects precisely to grasp them quickly and reliable. Therefore, classic image-based methods for object localization are not always applicable, as they may experience bad localization rates in large scenes or long runtimes at high camera resolutions. The focus of this paper is to provide a hybrid approach for the localization of objects in a scene by using a deep neural network for classification and segmentation of objects and classic shape-based matching for exact localization in the detected region of interest. We use Mask R-CNN as a deep neural network pixel based classifier and Halcon as an example for a well-known classic image processing program. Last but not least the solution is evaluated in comparison to only using Halcon by evaluating the performance regarding execution time and reliability of the object pose localization.