András Horváth

Grouping vehicles with similar paths is important for traffic analysis, but camera-image trajectories are distorted by perspective, and unsupervised clustering does not directly reveal how many traffic movements should be expected. This paper presents Homography-Guided Semantic Maneuver Graph Trajectory Clustering (HG-SMG-TC), a model-selection framework that uses a lightweight homography to estimate intersection structure while retaining isotropically normalized camera coordinates for clustering. Entry and exit endpoint groups are consolidated into physical approach-level groups, and their supported origin–destination relationships form a maneuver graph. Bootstrap resampling converts this structure into an interval for the expected number of observed movements, which guides clustering model selection. The method is evaluated on 67,029 vehicle trajectories from five urban intersection scenes in the Traffic Node Video Dataset, using separate target-estimation, model-selection, and independent-test recording blocks. Independent polygon-rule reference labels cover 89.6–98.4% of test trajectories. HG-SMG-TC reduces mean target-count error from 3.20 for untargeted selection and 2.53 for the point-target variant to 2.13, while achieving an adjusted Rand index of 0.734 and normalized mutual information of 0.839. The results show that the proposed semantic maneuver prior improves target alignment and provides a reproducible way to guide unsupervised trajectory clustering, while retaining explicit trade-offs across evaluation metrics.

Our study is motivated by the growing interest in intelligent transportation systems and computer vision-based approaches to traffic analysis. These advanced techniques can provide valuable information on traffic patterns, vehicle behaviours, and safety hazards at intersections, which are critical components of urban transportation networks. From the artificial intelligence perspective, the paper contributes a systematic comparative analysis of trajectory-based prediction methods using a complete detection–tracking–clustering pipeline and a structured feature-engineering framework. The study evaluates four classical machine learning models and twelve feature-vector configurations, providing insights into the accuracy, computational efficiency, and resource requirements of each. From the engineering application perspective, the proposed method supports early prediction of vehicle exit directions in real-world intersections using short two-second trajectory segments. This enables practical integration into real-time traffic monitoring and intelligent transportation systems. We use the You Only Look Once version 7 (YOLOv7) detector for vehicle detection and Deep Simple Online and Realtime Tracking (DeepSORT) for object tracking, followed by unsupervised trajectory clustering using the Ordering Points To Identify the Clustering Structure (OPTICS) algorithm. Our results show that a Multi-Layer Perceptron (MLP) combined with a compact six-dimensional feature vector provides the best balance between predictive performance and computational cost. We also introduce a graphical tool for visualising prediction outcomes directly on video frames.

The paper investigates disparities in food waste generation across European Union countries between 2020 and 2022, focusing on spatial and sustainability dimensions. It utilizes data for six key food waste parameters and a broad range of environmental, social and economic indicators. A combination of statistical methods, including correlation analysis, cluster analysis and Principal Component Analysis, uncovers multivariate patterns and identifies groups of countries with similar food waste characteristics and related factors. The paper highlights the temporal and spatial dynamics of food waste over the three-year period, particularly in light of the COVID-19 pandemic. While the total volume of food waste remained relatively stable across the EU, notable shifts occurred in waste sources. Household food waste peaked in 2021, likely due to increased time spent at home during pandemic-related lockdowns. Conversely, waste from retail, restaurants and food service sectors showed a consistent increase. The paper identifies non-trivial correlations between food waste and socio-economic variables, suggesting that differences in food waste generation across EU countries are influenced by a complex interplay of factors, including policy effectiveness, cultural practices, consumer behaviour and economic conditions. This comprehensive analysis of food waste patterns across EU countries and over time offers valuable insights for policymakers aiming to reduce waste and promote sustainability.

Az Intelligens Közlekedési Rendszerek (ITS) dinamikusan fejlődő és széles körben kutatott terület manapság. Nagyon szerteágazó tudományterület, amelynek fontos része a forgalmi dinamika elemzés, forgalomszámlálás és a forgalomi előrejelzések. Ebben a cikkben egy adaptív forgalomi statisztika generáló eljárást mutatunk be. A módszer álló megfigyelő szemszögéből készült videófelvételeken YOLOv7 objektumdetektorral detektálja a közlekedő járműveket és azonosítja azok kategóriáját. A videószekvenciákon elvégzett detektálásokat az általunk bemutatott módszer DeepSort követő algoritmus segítségével kapcsolja össze, aminek eredményeként előállnak az elhaladó járművek trajektóriái. A trajektóriák klaszterekbe sorolását OPTICS clustering eljárással végezzük el, aminek paraméterezésével szabályozzuk a létrejövő statisztika részletességét, ezáltal lehetővé válik akár utca-finomságú vagy sávszintű forgalmi adatok generálása ugyanabból a felvételből. Jelen cikkünkben négy különböző helyszínen készült felvételen teszteltük az eljárásunkat, aminek eredményeként nemcsak az egyes irányok tetszőleges felbontású forgalmi statisztikáit állíthatjuk elő, hanem azok időszakok szerinti és útirány szerinti normalizáltjait is. A normalizált statisztikai adatok megmutatják a forgalom szempontjából frekventált időszakokat és útvonalakat. A cikk végén gyakorlati hasznosíthatóságra irányuló példákon keresztül mutatjuk be a módszer teljesítőképességét.

Multimedia compression is an essential data processing procedure to any Content Delivery Network (CDN). Autoencoder is a data driven model which compresses the high dimensional inputs into a low dimensional latent representation creating a bottleneck impelling the model to learn a compression which is able to reconstruct the input with minimum error. The model searches for patterns in the data to reduce the input dimensionality fitting the input through the bottleneck while minimizing the reconstruction loss. Conventional autoencoder's latent representation exploits intra-sample correlation without taking advantage of the inter-samples correlation. Our model, Multi-instance Conditional Autoencoder, compresses many files to comprise one latent representation, and then restore each file individually in the decoding phase. We achieved higher compression rate compared to traditional autoencoder in case of audio files, Javanese and VoxCeleb datasets, where data correlation is sufficiently high.

Afterimages are common and frequent perceptual phenomena of everyday life. When looking into a bright light source compared to environment and suddenly turning the eyes away from it, the illusionary image of the light source remains in perception, for a while. A subordinate to afterimages defined in our research is virtual colour perception that, according to our interpretation is an unusual and intense temporary colour perception provoked by quick colour change in the incident light. Modelling virtual colour perception can be useful in the field of visual design, for example in dynamic advertising. A computational kinetic model established for simulating and quantifying virtual colour perception was developed in our previous work and was validated with preliminary tests involving 20 observers. Validation test results were contestable due to the high variance (about 50 RSD%) of duration and intensity of afterimage results at each incident light colour change. Thus, in our current work, validation on one-observer tests was performed to minimize the variance of results. Further, these validation tests were run on two liquid crystal display (LCD) types: a nowadays frequently present In-Plane Switching (IPS) display providing vivid colours, and an older LCD-TN (twisted nematic) display.