We propose an approach to estimate 3D human pose in real world units from a single RGBD image and show that it exceeds performance of monocular 3D pose estimation approaches from color as well as pose estimation exclusively from depth. Our approach builds on robust human keypoint detectors for color images and incorporates depth for lifting into 3D. We combine the system with our learning from demonstration framework to instruct a service robot without the need of markers. Experiments in real world settings demonstrate that our approach enables a PR2 robot to imitate manipulation actions observed from a human teacher.

No abstract

No abstract

Microcracks in silicon wafers reduce the strength of the wafers and can lead to critical failure within the solar-cell production. Both detection of the microcracks and their impact on fracture strength of the wafers are addressed within this study. To improve the accuracy of the crack detection in photoluminescence (PL) and infrared transmission (IR) images of as-cut wafers, we introduce a pattern recognition approach based on local descriptors and support-vector classification. The learning model requires a set of labeled data generated by an artificial insertion of cracks. Within this evaluation, the algorithm detects 81% of the cracks for PL-images and 98% for IR-images at precision rates above 98% in each case, which outperforms the quality of pure IR-intensitybased crack-detection systems with a hit-rate of 65% at a precision of 59%. The proposed algorithm may be combined with the images of the grain structure to avoid the confusion of cracks and grain boundaries. Moreover, the comprehensive set of wafers allows the impact of crack morphology on wafer strength to be investigated. Despite complex crack morphologies, the theoretically expected dependence between crack length and fracture strength is confirmed. Therefore, sorting criteria are derived to rate the cracks with respect to the expected fracture strength of the wafer based on the measured crack length only.

Reliable object grasping is a crucial capability for autonomous robots. However, many existing grasping approaches focus on general clutter removal without explicitly modeling objects and thus only relying on the visible local geometry. We introduce CenterGrasp, a novel framework that combines object awareness and holistic grasping. CenterGrasp learns a general object prior by encoding shapes and valid grasps in a continuous latent space. It consists of an RGB-D image encoder that leverages recent advances to detect objects and infer their pose and latent code, and a decoder to predict shape and grasps for each object in the scene. We perform extensive experiments on simulated as well as real-world cluttered scenes and demonstrate strong scene reconstruction and 6-DoF grasppose estimation performance. Compared to the state of the art, CenterGrasp achieves an improvement of 38.5 mm in shape reconstruction and 33 percentage points on average in grasp success. We make the code and trained models publicly available at http://centergrasp.cs.uni-freiburg.de.