OPUS - Publikationenserver der Technischen Hochschule Nürnberg Georg Simon Ohm
Not a member yet
2332 research outputs found
Sort by
Incoherent OFDR System for Distances up to 200 m SI-POF
No full textWe report on a proposal for an incoherent optical frequency domain reflectometry (I-OFDR) for 1-mm core diameter PMMA SI-POF. An avalanche photodiode receiver and optimized transmitter based on a 515-nm laser diode allow measurement up to 200 m SI-POF. The measurements have been carried out using a vector network analyzer; the subsequent data processing has been implemented in MATLAB
AI-based User Emotion Recognition from Interaction Data : Challenges and Guidelines for Training Data Creation
No full textArtificial Intelligence (AI) is a rising topic in the field of emotion recognition, e.g., from facial expressions. However, existing methods often require to be performed in staged set ups and are obtrusive by gathering additional data. Especially, collecting video data includes a high data protection risk. Our approach is to provide an unobtrusive emotion recognition tool based on Keystroke, Mouse and Touchscreen (KMT) data. Recently, we published a data set for emotion recognition from keystroke and mouse interaction data. In this paper, we present the challenges we faced during the creation of the data set. This covers collecting User Interface (UI) data as well as emotional ground truth data. For each of seven mentioned challenges, we provide our solutions as well as guidelines for other researchers to prevent them. The challenges include possible issues with recorded data as well as issues of automated facial coding engines. We provide a possible approach for manual facial coding and describe aspects attention should be paid to. Furthermore, we indicate issues when using different software tools to collect the data. The paper aims to help other researchers by providing insights and a guideline for the creation of the data set. We make these insights available for other researchers who want to create similar data sets or who want to expand ours. By sharing our insights, we aim to improve the reproducibility of AI training data creation and AI-based emotion recognition from user interaction data
RIV-SLAM: Radar-Inertial-Velocity optimization based graph SLAM
No full text4D imaging radars, commonly known as 4D radars, deliver comprehensive point cloud data that encapsulates range, azimuth, elevation, and Doppler velocity information even in harsh environmental conditions, such as
rain, snow, smoke, and fog. However, 4D radar data also suffers from high noise and sparsity, which poses great challenges for SLAM applications. This paper presents RIV-SLAM, a complete radar-inertial-velocity optimization-based graph SLAM system designed to exploit the full potential of 4D imaging radar technology. RIV-SLAM consists of four integral components: front-end, loop closure, IMU pre-integration and
graph optimization, each optimized to effectively leverage the unique attributes of radar data and tightly coupled with IMU data. This is also the first SLAM system known to us that outputs an optimized ego velocity. This capability ensures reliable ego motion estimation under extreme conditions (e.g., wheel odometry fails). Furthermore, we develop a new
ground extraction approach, specifically adapted for the 4D imaging radar, which substantially improves the system’s z-axis accuracy. Comprehensive evaluations of the RIV-SLAM system on a variety of datasets demonstrate its superior performance, significantly surpassing existing state-of-the-art Radar-SLAM frameworks. The code of RIV-SLAM will be released at: RIV-
SLA
Mehr als nur Antworten auf Papier
No full textWie digitale Recherchekompetenzen eingeübt und geprüft werden können, zeigt die Technische Hochschule Nürnberg am Beispiel einer Prüfung in der Sozialen Arbei
A Multidisciplinary Approach to AI-based self-motivated Learning and Teaching with Large Language Models
No full textWe present a learning experience platform that uses machine learning methods to support students and lecturers in self-motivated online learning and teaching processes. The platform is being developed as an agile open-source collaborative project supported by multiple universities and partners. The development is guided didactically, reviewed, and scientifically evaluated in several cycles. Transparency, data protection and the copyright compliant use of the system is a central part of the project. The system further employs large language models (LLMs). Due to privacy concerns, we utilize locally hosted LLM instances and explicitly do not rely on available cloud products. Students and lecturers can interact with an LLM-based chatbot in the current prototype. The AI-generated outputs contain cross-references to the current educational video’s context, indicating if sections are based on the lectures context or world knowledge. We present the prototype and results of our qualitative evaluation from the perspective of lecturers and students
Niedrigschwellige onlinebasierte Suchtberatung in Fußballfanszenen
No full textZusammenfassungIm Rahmen des Projektes SubFAN („Substanzkonsum in Fußballfanszenen“) wird ein niedrigschwelliges onlinebasiertes Beratungsportal entwickelt, das sich an jugendliche und junge erwachsene Fußballfans richtet. Das Projekt integriert Aspekte der Sucht- und Gewaltprävention und orientiert sich an der spezifischen Lebensrealität der Zielgruppe. Durch die Kooperation mit zehn sozialpädagogischen Fanprojekten wird eine Schnittstelle zwischen Jugend- und Suchthilfe geschaffen. Im Rahmen der Begleitforschung werden zudem die Hintergründe und Dynamiken von Substanzkonsum und Gewalterfahrungen im Fußballkontext analysiert
Parallax-Aware Image Stitching Based on Homographic Decomposition
No full textImage stitching plays a crucial role for various computer vision applications, like panoramic photography, video production, medical imaging and satellite imagery. It makes it possible to align two images captured at different views onto a single image with a wider field of view. However, for 3D scenes with high depth complexity and images captured from two different positions, the resulting image pair may exhibit significant parallaxes. Stitching images with multiple or large apparent motion shifts remains a challenging task, and existing methods often fail in such cases. In this paper, a novel image stitching pipeline is introduced, addressing the aforementioned challenge: First, iterative dense feature matching is performed, which results in a multi-homography decomposition. Then, this output is used to compute a per-pixel multidimensional weight map of the estimated homographies for image alignment via weighted warping. Additionally, the homographic image space decomposition is exploited using combinatorial analysis to identify parallaxes, resulting in a parallax-aware overlapping region: Parallax-free overlapping areas only require weighted warping and blending. For parallax areas, these operations are omitted to avoid ghosting artifacts. Instead, histogram- and mask-based color mapping is performed to ensure visual color consistency. The presented experiments demonstrate that the proposed method provides superior results regarding precision and handling of parallaxes
André Schülke (2023): Phänomene regressiver Nachrichtenrezeption. Medienethische Perspektiven auf die Konjunktur nichtrationaler mentaler Modelle der Realität.
No full textMultimodal Multimedia Event Argument Extraction with Unified Template Filling
No full textWith the advancement of multimedia technologies, news documents and user-generated content are often represented as multiple modalities, making Multimedia Event Extraction (MEE) an increasingly important challenge.
However, recent MEE methods employ weak alignment strategies and data augmentation with simple classification models, which ignore the capabilities of natural language-formulated event templates for the challenging Event Argument Extraction (EAE) task. In this work, we focus on EAE and address this issue by introducing a unified template filling model that connects the textual and visual modalities via textual prompts. This approach enables the exploitation of cross-ontology transfer and the incorporation of event-specific semantics. Experiments on the M2E2 benchmark demonstrate the effectiveness of our approach. Our system surpasses the current SOTA on textual EAE by +7% F1, and performs generally better than the second-best systems for multimedia EAE