Collisions on inland waters frequently occur due to the absence of clearly defined navigation guidance. To prevent such accidents, optical sensors such as stereo cameras can be employed to detect obstacles at an early stage, enabling timely warnings to vessel operators or the planning of collision avoidance trajectories. In this context, the paper presents a neural network for object detection on inland waters, designed to address the specific challenges of waterborne vehicles, including strong ego-motion and contextual cues like the shoreline appearing in the background. The proposed network leverages a plane sweep approach to integrate multiple views of a scene and predict object locations in bird’s-eye view (BEV) coordinates. Its ability to incorporate more than two camera perspectives is demonstrated using the KITTI dataset. On real-world inland water data, the network is evaluated against both a traditional maritime stereo-based method and a learning-based stereo detector from autonomous driving, showing consistently higher mean average precision. The code is available at https://github.com/dionysos4/IWOD.

Autonomous navigation on inland waters requires an accurate understanding of the environment in order to react to possible obstacles. Deep learning is a promising technique to detect obstacles robustly. However, supervised deep learning models require large data-sets to adjust their weights and to generalize to unseen data. Therefore, we equipped our research vessel with a laser scanner and a stereo camera to record a novel obstacle detection data-set for inland waters. We annotated 1974 stereo images and lidar point clouds with 3d bounding boxes. Furthermore, we provide an initial approach and a suitable metric2to compare the results on the test data-set. The data-set is publicly available3and seeks to make a contribution towards increasing the safety on inland waters.

One-class incremental learning is a special case of class-incremental learning, where only a single novel class is incrementally added to an existing classifier instead of multiple classes. This case is relevant in industrial defect detection scenarios, where novel defects usually appear during operation. Existing rolled-out classifiers must be updated incrementally in this scenario with only a few novel examples. In addition, it is often required that the base classifier must not be altered due to approval and warranty restrictions. While simple finetuning often gives the best performance across old and new classes, it comes with the drawback of potentially losing performance on the base classes (catastrophic forgetting [1]). Simple prototype approaches [2] work without changing existing weights and perform very well when the classes are well separated but fail dramatically when not. In theory, null-space training (NSCL) [3] should retain the basis classifier entirely, as parameter updates are restricted to the null space of the network with respect to existing classes. However, as we show, this technique promotes overfitting in the case of one-class incremental learning. In our experiments, we found that unconstrained weight growth in null space is the underlying issue, leading us to propose a regularization term (R-NSCL) that penalizes the magnitude of amplification. The regularization term is added to the standard classification loss and stabilizes null-space training in the one-class scenario by counteracting overfitting. We test the method’s capabilities on two industrial datasets, namely AITEX and MVTec, and compare the performance to state-of-the-art algorithms for class-incremental learning.

Motion estimation is an essential element for autonomous vessels. It is used e.g. for lidar motion compensation as well as mapping and detection tasks in a maritime environment. Because the use of gyroscopes is not reliable and a high performance inertial measurement unit is quite expensive, we present an approach for visual pitch and roll estimation that utilizes a convolutional neural network for water segmentation, a stereo system for reconstruction and simple geometry to estimate pitch and roll. The algorithm is validated on a novel, publicly available dataset22https://git.ios.htwg-konstanz.de/dgriesse/constance_orientation_dataset/archive/main/constance_orientation_dataset-main.zip recorded at Lake Constance. Our experiments show that the pitch and roll estimator provides accurate results in comparison to an Xsens IMU sensor. We can further improve the pitch and roll estimation by sensor fusion with a gyroscope. The algorithm is available in its implementation as a ROS node33https://github.com/dionysos4/water_surface_detector.

We are interested in computing a mini-batch-capable end-to-end algorithm to identify statistically independent components (ICA) in large scale and high-dimensional datasets. Current algorithms typically rely on pre-whitened data and do not integrate the two procedures of whitening and ICA estimation. Our online approach estimates a whitening and a rotation matrix with stochastic gradient descent on centered or uncentered data. We show that this can be done efficiently by combining Batch Karhunen-Löwe-Transformation [1] with Lie group techniques. Our algorithm is recursion-free and can be organized as feed-forward neural network which makes the use of GPU acceleration straight-forward. Because of the very fast convergence of Batch KLT, the gradient descent in the Lie group of orthogonal matrices stabilizes quickly. The optimization is further enhanced by integrating ADAM [2], an improved stochastic gradient descent (SGD) technique from the field of deep learning. We test the scaling capabilities by computing the independent components of the well-known ImageNet challenge (144 GB). Due to its robustness with respect to batch and step size, our approach can be used as a drop-in replacement for standard ICA algorithms where memory is a limiting factor.

In this paper we present a method using deep learning to compute parametrizations for B-spline curve approximation. Existing methods consider the computation of parametric values and a knot vector as separate problems. We propose to train interdependent deep neural networks to predict parametric values and knots. We show that it is possible to include B-spline curve approximation directly into the neural network architecture. The resulting parametrizations yield tight approximations and are able to outperform state-of-the-art methods.

In reverse engineering and computer-aided design (CAD) applications point cloud data is usually manually scanned, reconstructed, and post-processed in separated steps. When point cloud data resulting from a scanning process do not satisfy certain necessary reconstruction requirements, one must perform scanning again to enable proper reconstruction. On-line reconstruction of 3d geometry allows one to generate and update a CAD reconstruction on-line during the scanning process with an hand-held laser scanner. Thus, regions where the scanned data is insufficient for the reconstruction are detected on the fly to allow an immediate correction and improvement of the scanned data. This enables the operator to focus on critical regions in the scanned data to improve the reconstruction quality. We present an on-line segmentation and on-line reconstruction of basic geometric primitives. The presented methods allow for a real-time processing of a point stream. They utilize data structures that can be updated at any time when additional data from the stream has to be processed. This data is used to complete and improve the segmentation and reconstruction during the scanning process.

Abstract We present an adaptive subdivision scheme for unstructured tetrahedral meshes inspired by the ‐subdivision scheme for triangular meshes. Existing tetrahedral subdivision schemes do not support adaptive refinement and have traditionally been driven by the need to generate smooth three‐dimensional deformations of solids. These schemes use edge bisections to subdivide tetrahedra, which generates octahedra in addition to tetrahedra. To split octahedra into tetrahedra one routinely chooses a direction for the diagonals for the subdivision step. We propose a new topology‐based refinement operator that generates only tetrahedra and supports adaptive refinement. Our tetrahedral subdivision algorithm is motivated by the need to have one representation for the modeling, the simulation and the visualization and so to bridge the gap between CAD and CAE. Our subdivision algorithm design emphasizes on geometric quality of the tetrahedral meshes, local and adaptive refinement operations, and preservation of sharp geometric features on the boundary and in the interior of the physical domain.

Abstract We present a volumetric iso‐geometric finite element analysis based on Catmull‐Clark solids. This concept allows one to use the same representation for the modeling, the physical simulation, and the visualization, which optimizes the design process and narrows the gap between CAD and CAE. In our method the boundary of the solid model is a Catmull‐Clark surface with optional corners and creases to support the modeling phase. The crucial point in the simulation phase is the need to perform efficient integration for the elements. We propose a method similar to the standard subdivision surface evaluation technique, such that numerical quadrature can be used. Experiments show that our approach converges faster than methods based on tri‐linear and tri‐quadratic elements. However, the topological structure of Catmull‐Clark elements is as simple as the structure of linear elements. Furthermore, the Catmull‐Clark elements we use are C2‐continuous on the boundary and in the interior except for irregular vertices and edges.

In this article, we present regularly parametrized G k free-form spline surfaces that extend box and half-box splines over regular triangular grids. The polynomial degree of these splines is max{4 k + 1, ⌈3k/2 + 1⌉r}, where r ∈ ℕ can be chosen arbitrarily and determines the flexibility at extraordinary points. The G k splines presented in this article depend crucially on low-degree (re-)parametrizations of piecewise polynomial hole fillings. The explicit construction of such parametrizations forms the core of this work and we present two classes of singular and regular parametrizations. Also, we show how to build box and half-box spline surfaces of arbitrarily high smoothness with holes bounded by only n patches, in principle.

The authors improve the butterfly and Loop's (1987) algorithm. As a result they obtain subdivision algorithms for triangular nets which can be used to generate G/sup 1/- and G/sup 2/-surfaces, respectively.