Manual waste monitoring is a labor-intensive process that can lead to unnecessary trips and overflowing waste bins due to fixed inspection schedules. Existing automated systems rely on static sensors installed inside each bin, which are difficult to scale and have not yet gained wider acceptance. Micro aerial vehicles (MAVs) can address this problem by carrying the necessary sensors to different locations as needed. This paper presents an MAV-based solution for automated waste bin monitoring. RGB images captured by the MAV are processed by a CNN to estimate fill levels without the need for in-bin sensors. The CNN is trained on a custom dataset of simulated waste bins and evaluated with respect to dataset performance and in a real application scenario. A mobile application enables operators to configure bin locations and monitor the process in real time. The MAV is controlled by a reinforcement learning policy and autonomously navigates to each bin location. The evaluation demonstrates that the CNN achieves reliable performance on both simulated and real images, and that the integrated system autonomously completes full inspection cycles. Overall, the proposed system offers a scalable and cost-efficient alternative for sustainable waste management.

In spatially constrained environments, such as warehouses or industrial workspaces, users often face difficulties in defining virtual regions due to occlusions, physical barriers, or limited accessibility. This paper presents a multi-perspective Augmented Reality (AR) system designed to support the precise placement of 3D virtual borders in such scenarios. The approach integrates spatially aligned remote camera perspectives into a mobile AR application, allowing users to view and interact with virtual content from otherwise unreachable positions. A loosely coupled system architecture enables dynamic integration and removal of remote cameras, ensuring scalability and adaptability to diverse setups. We evaluate the system in a user study (N=17), assess its impact on physical and cognitive workload and to analyze the usage and effect of multiple perspectives during virtual object manipulation in constrained environments. Participants reported improved spatial understanding and ease of interaction, though occasional misplacement errors occurred when relying solely on static views. These findings suggest that integrating additional perspectives into AR interfaces can effectively enhance interaction in complex and constrained environments.

Robots are increasingly finding their way into home environments, where they can assist with household tasks like vacuuming or surveilling. While the robots can navigate on their own, users might not want them to go everywhere or not in a specific way. For example, users might not want a drone to fly over a table where important letters and the newspaper are stored, even though it is the shortest path to the goal. Therefore, an application is required, that is easy to learn and to apply even for inexperienced users.In this paper, we present a framework that uses a tablet as augmented reality (AR) device to modify a robot’s movement space in 3D. A user can define virtual borders in the real world with the tablet and add them to a map, changing the navigational behavior of the robot. The framework is evaluated by a user study with inexperienced participants that verifies our approach. Further analyses show, that even complex scenarios can be covered with our framework.

Abstract In this work, we focus on outdoor lighting estimation by aggregating individual noisy estimates from images, exploiting the rich image information from wide-angle cameras and/or temporal image sequences. Photographs inherently encode information about the lighting of the scene in the form of shading and shadows. Recovering the lighting is an inverse rendering problem and as that ill-posed. Recent research based on deep neural networks has shown promising results for estimating light from a single image, but with shortcomings in robustness. We tackle this problem by combining lighting estimates from several image views sampled in the angular and temporal domains of an image sequence. For this task, we introduce a transformer architecture that is trained in an end-2-end fashion without any statistical post-processing as required by previous work. Thereby, we propose a positional encoding that takes into account camera alignment and ego-motion estimation to globally register the individual estimates when computing attention between visual words. We show that our method leads to improved lighting estimation while requiring fewer hyperparameters compared to the state of the art.

This erratum aims to correct errors in the sections 1, 3, and 5 of Lee et al. (2022).Some of the texts in these sections were reproduced in non-final form.It resulted in omissions of several major extensions that are made during the revision process.Figures and Tables are not affected.

In this work, three different algorithms are examined for the evacuation of smart buildings, in particular a Static Evacuation Algorithm (SEA), a Dynamic Evacuation Algorithm (DEA) and a Fire Dynamic Algorithm (FDA). The Static Evacuation Algorithm represents the shortest path out of the building to a safe area. The Dynamic Algorithm calculates the optimal escape route based on the current position and position of a detected fire hazard. The Fire Dynamic escape route predicts how fast the fire will spread and includes this into the calculation of the escape route. The used simulation environment is based on the cross-platform game engine Unity3D and a building model was created using the building plan of the main building at the Campus Minden of University of Applied Sciences Bielefeld, Germany. We found that our proposed FDA performed 31.64% better than the SEA and 23.8% better than the DEA in terms of the hazard area over a minimally longer distance.

Palynology, the study of pollen, is becoming the focus of attention in computer vision in recent years. Various proposed automated classification and segmentation methods have been evaluated on a number of data sets. However, as of 2021 most data sets are sparse; they either contain only a small number of pollen classes, images in total or are imbalanced overall. In this work, we explore the possibility of creating synthetic pollen grain images from less than 2,000 images per pollen class via a Generative Adversarial Network (GAN). For that purpose, we selected two distinct pollen classes from a state of the art pollen data set and evaluated the data set with and without synthetic data on a Convolutional Neural Network (CNN). The enriched data set performed better overall (+1.4%) and specifically for the two pollen classes (+2%). We also drastically reduced the no. of real images and were still able to achieve a score of 60% to 80%. The experiments show, that our synthesized pollen images are visually close to real-life pollen grains and can be used to enrich imbalanced data sets as an addition to traditional data augmentation methods.