The Passive Leg Raise (PLR) test serves as an essential diagnostic instrument for evaluating fluid responsiveness within intensive care units (ICUs). Nevertheless, its execution and interpretation are susceptible to errors resulting from procedural complexity and clinician workload. We introduce A-PLR, an activity-aware assistant integrated into a smart ICU setting to mitigate these challenges. The system employs computer vision and spatial activity recognition to autonomously identify PLR phases, ensure procedural compliance, and provide real-time decision support via an intuitive bedside interface. This paper reports a mixed-method evaluation of A-PLR, comprising both quantitative performance metrics and qualitative insights from clinical professionals (N = 30). Within the controlled experimental setting, the results indicate improved usability and procedural support, as well as indications of enhanced diagnostic assistance and reduced cognitive load when clinicians were supported by A-PLR. In addition, user feedback highlights the potential of activity-aware, AI-driven systems to support decision-making and streamline workflow integration in (simulated) critical care environments.

Magnetic surveys are a key tool in detecting buried objects such as unexploded ordnance (UXO), where dense magnetic maps must be reconstructed from sparsely sampled gradiometer data. We present a deep learning-based approach that outperforms classical interpolation methods in both accuracy and speed. Trained on synthetic magnetic fields simulating realistic UXO signatures and measurement noise, our modified U-Net with ResNet-34 encoding reconstructs high-resolution magnetic maps from sparse inputs. Compared to state-of-the-art gridding methods, our model achieves 3–5% higher reconstruction accuracy on average while operating up to 80× faster than SOTA algorithms, enabling more efficient and interpretable UXO detection in real-world survey conditions.

User interfaces are essential tools for commanding machines on classical 2D displays and in immersive virtual reality (VR) environments. However, the limitless design options of 3D user interfaces and naturalistic interactions present challenges. While user experience for interaction in 2D scenarios has been studied for decades and has led to standards and well-established best practices, little is known about the rational design and the resulting user experience of user interfaces in VR. In this paper, we evaluate two user interaction methods: menu-based interaction and gesture interaction. We implemented an exemplary VR application for watching video content to explore 2D and 3D interaction concepts in a VR environment and collected user ratings for our interaction methods. Our results indicate that the menu-based hand interactions are superior to gesture-based ones in terms of task effectiveness and user satisfaction on average. However, a detailed analysis shows that for a significant subset of interaction elements gesture-based interaction is on par or superior to menu-based interaction while maintaining the subjective notion of naturalness. This provides insights into how to design efficient and natural gesture-based user interaction.

Being self-represented through an avatar increases embodiment and the feeling of presence in virtual reality. Nevertheless, currently users in VR are typically represented only by their hands, as not enough tracking data is available for full body self-representation. In our use case of VR-based diagnostics and cognitive rehabilitation of stroke patients, we aim for full body self-representation in order to increase therapeutic effectivity of the treatment. To solve this problem, Inverse Kinematics (IK) can be used for pose estimation, where no tracking data is available. IK allows to minimize the use of hardware and efforts of patients and clinical staff and, at the same time, provides a full-body representation based only on positions of the VR-HMD and users hands as input. In some use cases tracking data from additional, visual full body tracking sensors can be used to estimate the position of the lower body joints. In this study, we evaluate existing IK-based pose estimators; find that VRIK from Final IK is the most suitable approach for the given use case; integrate VRIK in our VR-rehabilitation system; adapt VRIK to meet the use case requirements and conduct subjective tests to validate a significantly increased notion of embodiment and presence through full-body over hands-only self-representation.

Accurate and reliable gesture recognition is a central problem in human-computer interaction (HCI). Many applications that make use of gesture recognition call for mobile devices with reduced power consumption, weight and form factors. Recent advances in computer vision were particularly brought by deep neural networks and come at the cost of high computational complexity that hinders the employment on mobile devices. In this study, we evaluate the usability of a low-cost Raspberry Pi 4B amended by a Coral USB Accelerator, or a Neural Compute Stick 2, respectively, for low power real-time gesture recognition. To this end we evaluate the accuracy, inference time and power consumption for two different deep neural network-based recognition models and compare the results to other computer systems available as standard. Our experiments show that a combination of a Raspberry Pi 4B and Coral USB Accelerator allows for hand gesture recognition at frame rates of up to 30 frames per second at a power consumption of less than 5 Watts.

The continued advancement in user interfaces comes to the era of virtual reality that requires a better understanding of how users will interact with 3D buttons in mid-air. Although virtual reality owns high levels of expressiveness and demonstrates the ability to simulate the daily objects in the physical environment, the most fundamental issue of designing virtual buttons is surprisingly ignored. To this end, this paper presents four variants of virtual buttons, considering two design dimensions of key representations and multi-modal cues (audio, visual, haptic). We conduct two multi-metric assessments to evaluate the four virtual variants and the baselines of physical variants. Our results indicate that the 3D-lookalike buttons help users with more refined and subtle mid-air interactions (i.e. lesser press depth) when haptic cues are available; while the users with 2D-lookalike buttons unintuitively achieve better keystroke performance than the 3D counterparts. We summarize the findings, and accordingly, suggest the design choices of virtual reality buttons among the two proposed design dimensions.