Band 92, Heft 7-8 - Special Issue: Image Processing Forum – Forum Bildverarbeitung 2024; Guest Editors: Michael Heizmann and Thomas Längle

This paper investigates how the noise characteristics of synthetically generated camera images correspond to those of a real camera. We determine the photon transfer curve from a set of rendered images of a static scene. Furthermore, we present a method to identify the regions with high temporal noise, i.e., rendering noise, in synthetically generated data from a single rendered image. Finally, we present a strategy on how a parameterization of the rendering can be achieved that minimizes the noise while also minimizing the rendering runtime.

This paper provides a methodology to directly compare the performance of event-based and traditional area scan cameras. Event-based cameras offer high temporal resolution, wide dynamic range (120 dB), low power consumption, and a compressed output stream, making them attractive for robotics and computer vision in challenging environments. However, the lack of a standardized comparison method between the two types of sensors has made it difficult to evaluate their relative performance and decide which camera is more suitable for which application. Here, we show that a set of three performance parameters, namely event contrast detectability (ECD), latency or maximum event rate, and event detection efficiency (EDE), can be used to quantitatively compare the performance of area scan and event-based cameras. The theoretical investigations are verified by a comparison study with 3 event-based and 3 industry standard area scan sensors, including a high dynamic range SONY IMX490 sensor to match the dynamic range of event-based cameras. For this purpose, the dynamic range of the EMVA 1288 compliant test equipment was extended to 120 dB.

Deep neural networks excel in perception tasks such as semantic segmentation and monocular depth estimation, making them indispensable in safety-critical applications like autonomous driving and industrial inspection. However, they often suffer from overconfidence and poor explainability, especially for out-of-domain data. While uncertainty quantification has emerged as a promising solution to these challenges, multi-task settings have yet to be explored. In an effort to shed light on this, we evaluate Monte Carlo Dropout, Deep Sub-Ensembles, and Deep Ensembles for joint semantic segmentation and monocular depth estimation. Thereby, we reveal that Deep Ensembles stand out as the preferred choice, particularly in out-of-domain scenarios, and show the potential benefit of multi-task learning with regard to the uncertainty quality in comparison to solving both tasks separately. Additionally, we highlight the impact of employing different uncertainty thresholds to classify pixels as certain or uncertain, with the median uncertainty emerging as a robust default.

In this contribution small semantic segmentation CNNs are evaluated against traditional segmentation approaches and state-of-the-art segmentation CNNs. The CNNs are optimized for the implementation on frame grabber FPGAs. A dataset of industrial burner flames and a dataset of transparent plastic granules is used to assess the segmentation performance of the models. VisualApplets by Basler AG is used to implement the models on an FPGA. The implemented models reach foreground IoU values of up to 96.8 % on our plastic granule dataset with a resolution of 256 × 256 pixel. In comparison state-of-the-art models evaluated reach values ranging from 94.0 % to 98.0 %. For images upscaled to 512 × 512 pixels, the performance of our models decreases to 91.4 % for the best FPGA implemented one. The inference of a 512 × 512 pixel image takes roughly 1 ms with an FPGA. The latency between the start of an input line to the output of the line is 0.1–1.9 ms for streaming an 8,192 pixel wide image. Using a GPU, the inference times on our models range between 0.59 ms and 1.03 ms, while the fastest state-of-the-art model reaches 3.44 ms.

Exploring the potential of quantum hardware for enhancing classical and real-world applications is an ongoing challenge. This study evaluates the performance of quantum and quantum-inspired methods compared to classical models for crack segmentation. Using annotated grayscale image patches of concrete samples, we benchmark a classical mean Gaussian mixture technique, a quantum-inspired fermion-based method, Q-Seg – a quantum-annealing-based method, and a U-Net deep learning architecture. Our results indicate that quantum-inspired and quantum methods offer a promising alternative to image segmentation, particularly for complex crack patterns, and could be applied in near-future applications.

A significant challenge in laser drilling is the optimization of process parameters and drilling strategies to achieve high-quality holes. This is further complicated by the fact that quality assessment is a manual and time-consuming task. This paper presents a methodology designed to significantly reduce the manual effort required in optimizing parameters for single-pulse laser drilling of 0.3 mm thick stainless steel. The objective is to precisely drill holes with an entry diameter of 70 µm and an exit diameter of 20 µm, achieving high roundness. The features of the drilled holes were extracted automatically from the raw data. The outcomes were compared against manual measurements. Results indicate that the mean deviations between automated and manual measurements for both inlet and outlet diameters are less than 1.5 µm. Based on the results of the feature extraction, we employed a Bayesian optimization algorithm to efficiently explore the parameter space without the need for incorporating expert knowledge. The approach rapidly identified optimal drilling parameters after only a few iterations, significantly expediting the optimization process and considerably reducing manual labor.

Zusammenfassung In dieser Arbeit präsentieren wir eine in Deutschland auf öffentlichen Straßen erfolgreich erprobte Systemarchitektur, um robust Haltelinien zugeordnete Ampeln wahrzunehmen und den resultierenden Freigabezustand zu filtern. Ampeldetektionen werden mit Ampeln aus einer HD-Karte assoziiert, um sie der Haltelinie zuzuordnen die das Fahrzeug betrifft. Somit kann die in der Karte hinterlegte Beziehung zwischen Ampeln und Haltelinie genutzt werden, ohne auf eine fehleranfällige Rückprojektion angewiesen zu sein. Wir evaluieren das Gesamtsystem anhand ausgewählter Szenarien in Karlsruhe und Sindelfingen und zeigen damit die Einsatzbereitschaft in realen automatisierten Fahrzeugen.