Autonomous vehicles offer the potential to drastically decrease number and severity of road accidents. Most accidents occur due human inattention or wrong decisions, whose factors can be eliminated by autonomous vehicles. However, not all are avoidable through automation. Complying with law is always enough, there environmental problems (bad weather, surface, etc.) causing accidents, other actors (human drivers, pedestrians) making mistakes. These unexpected situations, real-time sensors...

Deep neural networks and deep learning are becoming important popular techniques in modern services applications. The training of these is computationally intensive, because the extreme number trainable parameters large amount samples. In this brief overview, current solutions aiming to speed up process via parallel distributed computation introduced. necessary components strategies described from low-level communication protocols high-level frameworks for learning. implementations with...

Template matching is a classic technique used in image processing for object detection. It based on multiple matrix-based calculations, where there are no dependencies partial results, so parallel solutions could be created. In this article two GPU implemented methods presented and compared to the CPU-based sequential solution.

Generally, template matching methods are very sensible to image orientation, rotation or the size of template. However, there known occasions, when they perform better then other keypoint-based techniques. This article demonstrates a massively parallelizable method multi-scaled matching. By implementing it sequentially on CPU and GPU with naive approach, runtimes can be measured during test multiple scale sizes. With use GPU, 16-times lower. These results indicate that is place for further...

Diabetes mellitus (DM) is a major health concern among children with the widespread adoption of advanced technologies. However, concerns are growing about transparency, replicability, biasedness, and overall validity artificial intelligence studies in medicine.

The Radon transform and the reverse formulas are mostly used in computer tomography to reconstruct original shape. projections of an image could be vision for object recognition, based on projection matching. By comparing projections, effective method defined. To understand behavior basic properties inspected, results several transformations analyzed.

The aim of the article is to briefly summarize main challenges on field multi-view computer vision, and by proposing available techniques, novel methods may be developed several fields. After a grouping systems with multiple cameras proposed, current are reviewed, for every field, disadvantages applications pointed out.

The matching of the visual representations two objects is a very important task in most computer vision applications.In special cases, when all look alike and only small differences occur, difficulty increases.In this paper, novel method for low-quality images rear-viewed vehicles proposed, using multi-directional image projection functions.For GPU-accelerated implementations, data-parallel algorithm introduced.It concluded, that use multiple directions with fixed resolution number...

Representation learning of images using deep neural networks have shown great results in classificational tasks. In case instance recognition, or object re-identification other approaches are used. Siamese architectured convolutional were the first approach to learn from semantic distances, and give similarity two inputs. Triplet apply triplet loss based on furthest positive closest negative pair. this paper we present a method multi-directional image projections as an initial transformation...

Low-level object matching can be done using projection signatures. In case of a large number projections, the algorithm has to deal with less significant slices. A trivial approach would do statistical analysis or apply machine learning determine features. To take adjacent values matrices into account, convolutional neural network should used. compare two matrices, Siamese-structure heads applied. this paper, an experiment is designed and implemented analyze performance Siamese Convolutional...

Metric embedding learning is a special form of supervised learning: instead regression or classification similarity value predicted based on embedded vector distance. To implement such behavior, first the Siamese architecture was introduced, where training two input samples, and transformation model seeks to minimize distance between same-category increase different samples. deal with problem overtraining, triplet loss introduced in 2015, considering three samples at step. Triplet networks...

One of the most challenging area diabetes research is to provide such automated insulin delivery systems – so called artificial pancreas that have robust and adaptive capabilities in a highly sophysticated way. I.e. they are able actions at beginning therapy satisfy requirements patients without knowing users daily lifestyle preferences however on short-term learn these patient specifics increase quality therapy. possible solution closed-loop self-learning features. In present study, we...

This paper introduces a novel method to calculate multi-directional projections of squared images, similar the Radon transformation. Image are often used as object signatures for detection, matching and tracking techniques in computer vision. The transformation provides fast solution these pixel intensity sums. proposed is based on trigonometric functions basic coordinate-geometry. implemented sequentially runtimes GPU-based implementation measured evaluated. analysis results indicate that...

Fault detection, which requires a lot of time and complexity, is one the most difficult tasks for cloud computing. In this research, we investigate utilization anytime algorithm in cloud-based fault detection techniques. Anytime algorithms are real-time decision-making that focus on time-deliberation. paper, basic concepts existing works as well its benefits usage analyzed. Various literature reviewed combined possible recognition deliberation. The paper includes details processing different...

The localisation of road accident hotspots (areas the public network, where number accidents is higher than expected based on average values) one main tasks safety experts. Analysing these locations can help to avoid further traffic and personal injuries (most all, fatal accidents). There are several already known methods in literature for this purpose paper contains a comparison between most frequently used ones: sliding-window method (using constant variable window size), 2D width height),...

Heatmaps are widely used by road safety engineers to visualise accident data (where heat refers the concentration of accidents). In subfield black spot localisation, high number similar candidate locations makes this method unreliable and difficult use. An additional process convert clusters heatmap into a list candidates results in an objective, comparable outcome. As we will show, commonly thresholding (considering pixels above given limit as parts spots) has several limitations. This...

Image based instance recognition is a difficult problem, in some cases even for the human eye. While latest developments computer vision—mostly driven by deep learning—have shown that high performance models classification or categorization can be engineered, problem of discriminating similar objects with low number samples remain challenging. Advances from multi-class are applied object matching problems, as feature extraction techniques same; nature-inspired multi-layered convolutional...

To simulate the behaviour of cell-level biological interactions, cells can be modeled as individual agents interacting with each other. identify effects parameter perturbation, a massive number simulations are usually required that feasible to carry out by parallel computation. In this article, multiprocessing scheme is defined reduce overall runtime. The relies on predicting expected computational cost input set, and scheduling them based Longest Processing Time (LPT) heuristic rule.

The exponential growth of online textual content across diverse domains has necessitated advanced methods for automated text classification. Large Language Models (LLMs) based on transformer architectures have shown significant success in this area, particularly natural language processing (NLP) tasks. However, general-purpose LLMs often struggle with domain-specific content, such as scientific texts, due to unique challenges like specialized vocabulary and imbalanced data. In study, we...