Open Access

Thomas Rosenberger

• The advent of autonomous and self-driving cranes represents a significant advancement in industrial automation. One critical prerequisites for achieving this long-term goal is the accurate and reliable detection of tools guided by ropes in real-world environments. Since the tool is suspended by ropes, the tool pose cannot be controlled directly. This master’s thesis addresses the challenges of pose estimation for rope-guided tools using point cloud measurements. The proposed algorithm utilizes constraints imposed by the crane kinematics and information extracted during the segmentation process to efficiently infer the pose of the hook, therefore enabling the use of the pose for decision making in real-time critical applications. RANSAC (Random Sample and Consensus) is deployed in the segmentation process to extract geometric primitives from the point cloud which represent the ropes and distinctive parts of the tool. Since the point cloud is often to sparse for feature matching a bounding box is used to estimate the initial position of the tool. Two different methods are presented to improve the initial pose. A computationally expensive method with a high level of confidence, integrating the ICP (Iterative Closest Point) algorithm is used as a benchmark. A linear Kalman filter is used in the second method which is real-time capable. The benchmark is then used to evaluate the real-time capable approach. The core contributions of this research lie in the innovative utilization of bounding boxes for pose estimation. The findings and methodologies presented herein constitute an advancement towards the realization of autonomous and self-driving cranes.