This work considers robot keypoint estimation on color images as a supervised machine learning task. We propose the use of probabilistically created renderings to overcome the lack of labeled real images. Rather than sampling from stationary distributions, our approach introduces a feedback mechanism that constantly adapts probability distributions according to current training progress. Initial results show, our approach achieves near-human-level accuracy on real images. Additionally, we demonstrate that feedback leads to fewer required training steps, while maintaining the same model quality on synthetic data sets.

For many industrial machine vision applications it is difficult to acquire good training data to deploy deep learning techniques. In this paper we propose a method based on probabilistic modelling and rendering to generate artificial images of carbon fiber fabrics. We deploy a convolutional neural network (CNN) to learn detection of fabric contours from artificially generated images. Our network largely follows the recently proposed U-Net architecture. We provide results for a set of real images taken under controlled lighting conditions. The method can easily be adapted to similar problems in quality control for composite parts.

Commodity RGB-D sensors capture color images along with dense pixel-wise depth information in real-time. Typical RGB-D sensors are provided with a factory calibration and exhibit erratic depth readings due to coarse calibration values, ageing and thermal influence effects. This limits their applicability in computer vision and robotics. We propose a novel method to accurately calibrate depth considering spatial and thermal influences jointly. Our work is based on Gaussian Process Regression in a four dimensional Cartesian and thermal domain. We propose to leverage modern GPUs for dense depth map correction in real-time. For reproducibility we make our dataset and source code publicly available.

In this paper, a robust, real-time object tracking approach capable of dealing with multiple symmetric and non-symmetric objects in a real-time requirement setting is proposed. The approach relies only on depth data to track multiple objects in a dynamic environment and uses random-forest based learning to deal with problems like object occlusion, motion-blur due to camera motion and clutter. We show that the relation between object motion and the corresponding change in its 3D point cloud data can be learned using only 6 random forests. A framework that unites object pose estimation and object pose tracking to efficiently track multiple objects in 3D space is presented. The approach is robust in tracking objects even in presence of motion blur that causes noisy depth data and is capable of real-time performance with 1.8ms per frame. The experimental evaluations demonstrate the performance of the approach against robustness, accuracy and speed and compare the approach with the state of the art. A publicly available dataset with real-world data is also provided for future benchmarking.