Refine
Year of publication
Document Type
- Bachelor Thesis (4)
- Doctoral Thesis (4)
- Master's Thesis (3)
- Diploma Thesis (2)
- Study Thesis (1)
Language
- English (14) (remove)
Keywords
- Action Recognition (1)
- Action Segmentation (1)
- Automatische Klassifikation (1)
- Befahrbarkeit (1)
- Bildanalyse (1)
- Bildverarbeitung (1)
- Coloskopie (1)
- Computer Vision (1)
- Darmpolyp (1)
- Deep Metric Learning (1)
Institute
The Web is an essential component of moving our society to the digital age. We use it for communication, shopping, and doing our work. Most user interaction in the Web happens with Web page interfaces. Thus, the usability and accessibility of Web page interfaces are relevant areas of research to make the Web more useful. Eye tracking is a tool that can be helpful in both areas, performing usability testing and improving accessibility. It can be used to understand users' attention on Web pages and to support usability experts in their decision-making process. Moreover, eye tracking can be used as an input method to control an interface. This is especially useful for people with motor impairment, who cannot use traditional input devices like mouse and keyboard. However, interfaces on Web pages become more and more complex due to dynamics, i.e., changing contents like animated menus and photo carousels. We need general approaches to comprehend dynamics on Web pages, allowing for efficient usability analysis and enjoyable interaction with eye tracking. In the first part of this thesis, we report our work on improving gaze-based analysis of dynamic Web pages. Eye tracking can be used to collect the gaze signals of users, who browse a Web site and its pages. The gaze signals show a usability expert what parts in the Web page interface have been read, glanced at, or skipped. The aggregation of gaze signals allows a usability expert insight into the users' attention on a high-level, before looking into individual behavior. For this, all gaze signals must be aligned to the interface as experienced by the users. However, the user experience is heavily influenced by changing contents, as these may cover a substantial portion of the screen. We delineate unique states in Web page interfaces including changing contents, such that gaze signals from multiple users can be aggregated correctly. In the second part of this thesis, we report our work on improving the gaze-based interaction with dynamic Web pages. Eye tracking can be used to retrieve gaze signals while a user operates a computer. The gaze signals may be interpreted as input controlling an interface. Nowadays, eye tracking as an input method is mostly used to emulate mouse and keyboard functionality, hindering an enjoyable user experience. There exist a few Web browser prototypes that directly interpret gaze signals for control, but they do not work on dynamic Web pages. We have developed a method to extract interaction elements like hyperlinks and text inputs efficiently on Web pages, including changing contents. We adapt the interaction with those elements for eye tracking as the input method, such that a user can conveniently browse the Web hands-free. Both parts of this thesis conclude with user-centered evaluations of our methods, assessing the improvements in the user experience for usability experts and people with motor impairment, respectively.
On the recognition of human activities and the evaluation of its imitation by robotic systems
(2023)
This thesis addresses the problem of action recognition through the analysis of human motion and the benchmarking of its imitation by robotic systems.
For our action recognition related approaches, we focus on presenting approaches that generalize well across different sensor modalities. We transform multivariate signal streams from various sensors to a common image representation. The action recognition problem on sequential multivariate signal streams can then be reduced to an image classification task for which we utilize recent advances in machine learning. We demonstrate the broad applicability of our approaches formulated as a supervised classification task for action recognition, a semi-supervised classification task for one-shot action recognition, modality fusion and temporal action segmentation.
For action classification, we use an EfficientNet Convolutional Neural Network (CNN) model to classify the image representations of various data modalities. Further, we present approaches for filtering and the fusion of various modalities on a representation level. We extend the approach to be applicable for semi-supervised classification and train a metric-learning model that encodes action similarity. During training, the encoder optimizes the distances in embedding space for self-, positive- and negative-pair similarities. The resulting encoder allows estimating action similarity by calculating distances in embedding space. At training time, no action classes from the test set are used.
Graph Convolutional Network (GCN) generalized the concept of CNNs to non-Euclidean data structures and showed great success for action recognition directly operating on spatio-temporal sequences like skeleton sequences. GCNs have recently shown state-of-the-art performance for skeleton-based action recognition but are currently widely neglected as the foundation for the fusion of various sensor modalities. We propose incorporating additional modalities, like inertial measurements or RGB features, into a skeleton-graph, by proposing fusion on two different dimensionality levels. On a channel dimension, modalities are fused by introducing additional node attributes. On a spatial dimension, additional nodes are incorporated into the skeleton-graph.
Transformer models showed excellent performance in the analysis of sequential data. We formulate the temporal action segmentation task as an object detection task and use a detection transformer model on our proposed motion image representations. Experiments for our action recognition related approaches are executed on large-scale publicly available datasets. Our approaches for action recognition for various modalities, action recognition by fusion of various modalities, and one-shot action recognition demonstrate state-of-the-art results on some datasets.
Finally, we present a hybrid imitation learning benchmark. The benchmark consists of a dataset, metrics, and a simulator integration. The dataset contains RGB-D image sequences of humans performing movements and executing manipulation tasks, as well as the corresponding ground truth. The RGB-D camera is calibrated against a motion-capturing system, and the resulting sequences serve as input for imitation learning approaches. The resulting policy is then executed in the simulated environment on different robots. We propose two metrics to assess the quality of the imitation. The trajectory metric gives insights into how close the execution was to the demonstration. The effect metric describes how close the final state was reached according to the demonstration. The Simitate benchmark can improve the comparability of imitation learning approaches.
Since the invention of U-net architecture in 2015, convolutional networks based on its encoder-decoder approach significantly improved results in image analysis challenges. It has been proven that such architectures can also be successfully applied in different domains by winning numerous championships in recent years. Also, the transfer learning technique created an opportunity to push state-of-the-art benchmarks to a higher level. Using this approach is beneficial for the medical domain, as collecting datasets is generally a difficult and expensive process.
In this thesis, we address the task of semantic segmentation with Deep Learning and make three main contributions and release experimental results that have practical value for medical imaging.
First, we evaluate the performance of four neural network architectures on the dataset of the cervical spine MRI scans. Second, we use transfer learning from models trained on the Imagenet dataset and compare it to randomly initialized networks. Third, we evaluate models trained on the bias field corrected and raw MRI data. All code to reproduce results is publicly available online.
This thesis addresses the problem of terrain classification in unstructured outdoor environments. Terrain classification includes the detection of obstacles and passable areas as well as the analysis of ground surfaces. A 3D laser range finder is used as primary sensor for perceiving the surroundings of the robot. First of all, a grid structure is introduced for data reduction. The chosen data representation allows for multi-sensor integration, e.g., cameras for color and texture information or further laser range finders for improved data density. Subsequently, features are computed for each terrain cell within the grid. Classification is performedrnwith a Markov random field for context-sensitivity and to compensate for sensor noise and varying data density within the grid. A Gibbs sampler is used for optimization and is parallelized on the CPU and GPU in order to achieve real-time performance. Dynamic obstacles are detected and tracked using different state-of-the-art approaches. The resulting information - where other traffic participants move and are going to move to - is used to perform inference in regions where the terrain surface is partially or completely invisible for the sensors. Algorithms are tested and validated on different autonomous robot platforms and the evaluation is carried out with human-annotated ground truth maps of millions of measurements. The terrain classification approach of this thesis proved reliable in all real-time scenarios and domains and yielded new insights. Furthermore, if combined with a path planning algorithm, it enables full autonomy for all kinds of wheeled outdoor robots in natural outdoor environments.
The goal of this minor thesis is to integrate a robotic arm into an existing robotics software. A robot built on top of this stack should be able to participate successfully RoboCup @Home league. The robot Lisa (Lisa is a service android) needs to manipulate objects, lifting them from shelves or handing them to people. Up to now, the only possibility to do this was a small gripper attached to the robot platform. A "Katana Linux Robot" of Swiss manufacturer Neuronics has been added to the robot for this thesis. This arm needs a driver software and path planner, so that the arm can reach its goal object "intelligently", avoiding obstacles and creating smooth, natural motions.
The automatic detection of position and orientation of subsea cables and pipelines in camera images enables underwater vehicles to make autonomous inspections. Plants like algae growing on top and nearby cables and pipelines however complicate their visual detection: the determination of the position via border detection followed by line extraction often fails. Probabilistic approaches are here superior to deterministic approaches. Through modeling probabilities it is possible to make assumptions on the state of the system even if the number of extracted features is small. This work introduces a new tracking system for cable/pipeline following in image sequences which is based on particle filters. Extensive experiments on realistic underwater videos show robustness and performance of this approach and demonstrate advantages over previous works.
Colonoscopy is the gold standard for the detection of colorectal polyps that can progress into cancer. In such an examination, physicians search for polyps in endoscopic images. Thereby polyps can be removed. To support experts with a computer-aided diagnosis system, the University of Koblenz-Landau currently makes some efforts in research different methods for automatic detection. Comparable to traditional pattern recognition systems, features are initially extracted and a classifier is trained on such data. Afterwards, unknown endoscopic images can be classified with the previously trained classifier. This thesis concentrates on the extension of the feature extraction module in the existing system. New detection methods are compared to existing techniques. Several features are implemented, incorporating Graylevel Co-occurrence Matrices, Local Binary Patterns and Discrte Wavelet Transform. Different modifications on those features are applied and evaaluated.
This thesis evaluates automated techniques to remove objects from an image and proposed several modifications for the specific application of removing a colour checker from structure dominated images. The selection of approaches covers the main research field of image inpainting as well as an approach used in medical image processing. Their results are investigated to disclose their applicability to removing objects from structure-intense images. The advantages and disadvantages discovered in the process are then used to propose several modifications for an adapted inpainting approach suitable for removing the colour checker.
Particle swarm optimization is an optimization technique based on simulation of the social behavior of swarms.
The goal of this thesis is to solve 6DOF local pose estimation using a modified particle swarm technique introduced by Khan et al. in 2010. Local pose estimation is achieved by using continuous depth and color data from a RGB-D sensor. Datasets are aquired from different camera poses and registered into a common model. Accuracy and computation time of the implementation is compared to state of the art algorithms and evaluated in different configurations.
This thesis addresses the automated identification and localization of a time-varying number of objects in a stream of sensor data. The problem is challenging due to its combinatorial nature: If the number of objects is unknown, the number of possible object trajectories grows exponentially with the number of observations. Random finite sets are a relatively new theory that has been developed to derive at principled and efficient approximations. It is based around set-valued random variables that contain an unknown number of elements which appear in arbitrary order and are themselves random. While extensively studied in theory, random finite sets have not yet become a leading paradigm in practical computer vision and robotics applications. This thesis explores random finite sets in visual tracking applications. The first method developed in this thesis combines set-valued recursive filtering with global optimization. The problem is approached in a min-cost flow network formulation, which has become a standard inference framework for multiple object tracking due to its efficiency and optimality. A main limitation of this formulation is a restriction to unary and pairwise cost terms. This circumstance makes integration of higher-order motion models challenging. The method developed in this thesis approaches this limitation by application of a Probability Hypothesis Density filter. The Probability Hypothesis Density filter was the first practically implemented state estimator based on random finite sets. It circumvents the combinatorial nature of data association itself by propagation of an object density measure that can be computed efficiently, without maintaining explicit trajectory hypotheses. In this work, the filter recursion is used to augment measurements with an additional hidden kinematic state to be used for construction of more informed flow network cost terms, e.g., based on linear motion models. The method is evaluated on public benchmarks where a considerate improvement is achieved compared to network flow formulations that are based on static features alone, such as distance between detections and appearance similarity. A second part of this thesis focuses on the related task of detecting and tracking a single robot operator in crowded environments. Different from the conventional multiple object tracking scenario, the tracked individual can leave the scene and later reappear after a longer period of absence. Therefore, a re-identification component is required that picks up the track on reentrance. Based on random finite sets, the Bernoulli filter is an optimal Bayes filter that provides a natural representation for this type of problem. In this work, it is shown how the Bernoulli filter can be combined with a Probability Hypothesis Density filter to track operator and non-operators simultaneously. The method is evaluated on a publicly available multiple object tracking dataset as well as on custom sequences that are specific to the targeted application. Experiments show reliable tracking in crowded scenes and robust re-identification after long term occlusion. Finally, a third part of this thesis focuses on appearance modeling as an essential aspect of any method that is applied to visual object tracking scenarios. Therefore, a feature representation that is robust to pose variations and changing lighting conditions is learned offline, before the actual tracking application. This thesis proposes a joint classification and metric learning objective where a deep convolutional neural network is trained to identify the individuals in the training set. At test time, the final classification layer can be stripped from the network and appearance similarity can be queried using cosine distance in representation space. This framework represents an alternative to direct metric learning objectives that have required sophisticated pair or triplet sampling strategies in the past. The method is evaluated on two large scale person re-identification datasets where competitive results are achieved overall. In particular, the proposed method better generalizes to the test set compared to a network trained with the well-established triplet loss.