Blekinge Institute of Technology
Not a member yet
13576 research outputs found
Sort by
Dynamisk ljussättning för virtuell produktion : Matchning av förgrund och bakgrund vid filmskapande i LED-baserades virtuella produktionsstudior
No full textThis article examines how new lighting methods can be used for filmmaking in virtual production andwhat impact they have on the visual result as well as on the workflow. Virtual production is a novel way of using digital real-time rendered backgrounds along with physical foreground elements. In this article, we interview experts in the field to understand the current situation. We learn that one of the primary difficulties of this production method is to match these fore-/background elements in order to sell the illusion of them existing in the same space. To explore this problem, we develop novel lighting solutions using dynamic possibilities possible through a real-time graphics engine, regularly used for virtual production. We then test and implement them in live studio environments and evaluate them together with industry experts. Finally we discuss the findings.Denna artikel undersöker hur nya ljussättningsmetoder kan användas för filmskapande inom virtuell produktion och hur de påverkar det visuella resultatet samt arbetsflödet. Virtuell produktion är ett nytt sätt att använda digitala realtidsbakgrunder tillsammans med fysiska förgrundselement. I denna artikel så intervjuar vi experter för att bättre förstå den nuvarande situationen. Vi lär oss att en av de stora svårigheterna med produktionsmetoden är att matcha dessa för-/bakgrundselement och på så sätt sälja illusionen att de existerar på samma plats. I syfte att utforska dessa problem så utvecklade vi nya dynamiska ljussättningsmetoder, möjliga genom en realtidsgrafikmotor. Vi testade och implementerade dessa sedan i produktionsstudiomiljöer, där de utvärderades tillsammans med experter. Slutligen så diskuterar vi resultaten
AI inom kommunal fysisk planering
No full textDetta kandidatarbete undersöker hur artificiell intelligens (AI) används och integreras i den kommunala fysiska planeringsverksamheten i Sverige. Studien syftar till att bidra med kunskap om hur svenska kommuner implementerar AI. Arbetet kombinerar dokumentstudier av offentliga styrdokument och projektbeskrivningar med semistrukturerade intervjuer genomförda med kommunala tjänstepersoner. Genom att tillämpa Rogers analytiska ramverk om organisatoriska innovationsprocesser analyseras varför svenska kommuner vill använda AI och vilka förväntningar samt farhågor som finns kring AI. Hur AI i dagsläget används, vad det finns för risker med att använda AI och vad för typ av resursutbyte som krävs. Resultatet visar att AI har flera användningsområden, däribland att effektivisera administrativa processer, förbättra beslutsunderlag och att vara ett stöd för kommunanställda. Arbetet har dock visat att den storskaliga AI-användningen ofta är beroende av externa samarbeten med företag medan den individuella AI-användningen kräver att kommunanställda utvecklar sin egen digitala kompetens. Studien visar även att den generella inställningen är både hoppfull och tveksam, samtidigt som läckande data, hallucinerande AI och transparensfrågor utgör hinder för implementering. AI betraktas av flera kommuner som ett potentiellt verktyg för ökad kvalitet och resurseffektivitet, men också som en teknik som kräver försiktighet, särskilt i relation till rättssäkerhet, transparens och personlig integritet. Arbetet bidrar till en ökad förståelse för hur digitala innovationer formas i kommunal verksamhet och understryker vikten av tvärsektoriellt lärande i framtida AI-satsningar
Synthetic Cloud and Shadow Generation for Deep Learning-Based Removal from Optical Satellite Images
No full textBackground: Optical satellite imagery is integral to various Earth observation initiatives, including environmental monitoring, urban planning, and disaster management. However, approximately 67% of the Earth’s surface is obscured by clouds and their shadows, which significantly hampers ground visibility and limits the effectiveness of this imagery. Traditional techniques for cloud removal typically depend on multi-temporal data, which is not always readily available, rendering these methods less reliable in regions characterized by persistent cloud cover. Objectives: This thesis endeavors to develop a comprehensive synthetic data generation framework, integrated with a deep learning architecture, aimed at reconstructing cloud-free satellite images. The primary objectives include: (1) the creation of realistic synthetic overlays of clouds and shadows for training purposes, (2) the design of an efficient model that accurately performs image reconstruction, and (3) the evaluation of the model’s performance in comparison to established generative adversarial networks. Methods: A novel Progressive Multi-scale Attention Autoencoder (PMAA) was designed to eliminate shadows and clouds from single-image inputs using attention refined processing (ARPD) at different scales. A dataset based on Perlin noise was developed for volumetric rendering to simulate varying atmospheric conditions. The model was trained and compared with CTGAN and STGAN baselines using different metrics of PSNR, SSIM, MSE, RMSE, MAE and Inference Time. Additional ablation experiments evaluated the contribution of architectural components to the design. Results: PMAA significantly outperformed baseline models, achieving a PSNR of 35.08 dB and an SSIM of 0.9673, indicating superior reconstruction quality and structural fidelity. It demonstrated robustness across scenarios with varying cloud densities and shadows, while remaining computationally efficient (average inference time of 0.0341s per image). The attention mechanism and progressive refinement were confirmed as key contributors through ablation analysis. Conclusions: This study confirms the effectiveness of synthetic training data and multi-scale attention in enhancing satellite image clarity under atmospheric occlusions. The PMAA framework sets a new benchmark for cloud and shadow removal, offering high-quality reconstructions with practical deployment potential in real world Earth observation applications
Improving a Dictionary-Based Minimum Forward Matching Tokenizer to Better Handle Word Segmentation of Japanese Sentences
No full textThis thesis addresses the challenge of accurately tokenizing Japanese sentences by improving a sole trader’s Python-based sentence-splitter system to better handle the splitting of Japanese sentences. Japanese poses unique difficulties during tokenization due to its agglutinative nature and due to the fact that spaces are not used to delimit words. We evaluate the baseline system, having a token lexicon of some 120 000 entries, against a Short Unit Word gold standard consisting of a sample of 10 000 sentences from Tatoeba, splitted using fugashi and the UniDic token lexicon. Precision, recall and F1-scores are calculated and an F1-score of 0.6537 is established for the baseline system. We enhance the token lexicon by (1) adding root verb and root adjective stems, (2) adding five additional verb stems that are used in UniDic and (3) incorporating 5 050 high-frequency tokens missing from the original lexicon. These additions raise the F1-score to 0.7235, demonstrating that while token lexicon expansion can yield some improvements in F1-score it is ultimately insufficient to reach the industry standard (96-99 %) without either pruning non-UniDic entries from the customer’s pre-existing token lexicon or by altering the phrase matching algorithm in the customer’s sentence-splitter system (which is currently a Minimum Forward Matching method). It is concluded that, while a base of Short Unit Words in the customer’s token lexicon is ultimately necessary, a move to a Long Unit Word standard may allow the customer to reach an industry standard, or somewhere close to it, while preserving the pre-existing data in the customer’s token lexicon
3D-poseuppskattning och tidsserieklassificering för att skilja mellan normala och utmattade tillstånd : Smartphone-baserad metod
No full textBackground: Subtle shifts in gait, such as slower steps or reduced joint angles,often signal fatigue but are difficult to detect without costly sensors or lab-grademotion-capture systems.Objectives: This thesis develops a video-based fatigue detection system using consumer smartphones, aiming to make motion analysis accessible. Specifically, we aimto:• Capture synchronized multi-angle videos using three smartphones;• Extract 2D keypoints with MMPose HRNet and reconstruct 3D poses usingSemGCN;• Compute kinematic features (joint velocities, accelerations, angles);• Train machine learning models (Conv1D-BiLSTM and RandomForest-GradientBoosting)to classify normal vs. fatigued gait and predict fatigue scores.Method: Volunteers walked normally and after fatiguing exercise while three smartphones recorded from different angles. We synchronized the video streams, extracted 2D keypoints using MMPose HRNet, and reconstructed 3D joint tracks withSemGCN. Kinematic features were calculated and segmented into 20-frame windows,feeding into a Conv1D-BiLSTM network and a RandomForest-GradientBoosting ensemble. Models were trained with 4-fold stratified cross-validation, early stopping,and learning-rate scheduling. Performance was evaluated via classification accuracyand mean absolute error (MAE) for fatigue scores, with robustness tested undervarying lighting and camera positions.Results: The Conv1D-BiLSTM model achieved 83.7% accuracy (±1.8%) and MAE= 2.6 on a 0–5 fatigue scale. The RandomForest-GradientBoosting ensemble achieved93.1% accuracy (±1.5%) and MAE = 1.9. Ablation studies showed kinematic features improved accuracy by up to 7%, and performance remained stable (within 7%)under lighting and angle variations.Conclusion: Using only smartphones and open-source tools, our pipeline reliablydetects fatigued gait and quantifies fatigue levels, enabling accessible health trackingfor occupational safety, sports, and rehabilitation.Keywords: 3D pose estimation, fatigue detection, MMPose HRNet, SemGCN, BiLSTM, Random Forest, smartphone video, kinematic features
Individanpassad spelifiering : Förbättrad inlärning i astronomi genom Big Five-modellen
No full textDenna studie undersöker hur spelifiering kan anpassas utifrån Big Five-personlighetsmodellen för att öka elevers motivation och lärande i fysik ämnen, med särskilt fokus på astronomi. Bakgrunden utgår från en problematik där undervisningen ofta upplevs som motiverande endast i begränsad omfattning, trots skolans ambition att möta elevers individuella behov. Genom litteraturstudier och en enkät riktad till verksamma lärare i årskurs 4–6 har studien kartlagt uppfattningar om spelifieringens potential och hur individanpassade digitala läromedel upplevs i undervisningen. Resultatet visar att lärare generellt ser spelifiering som ett lustfyllt komplement till traditionell undervisning, men också att det finns risker med spelelement som överbetonar tävling eller snabba belöningar. Vidare presenterar studien en spelbar prototyp av ett digitalt läromedel där spelmekanismer anpassats efter användarens personlighet genom ett förenklat Big Five-baserat test. Prototypen visualiserar ett pedagogiskt rymdäventyr med delmoment och kunskapstest kopplade till astronomiska fenomen. Studien drar slutsatsen att spelifiering med personlighets anpassning har potential att främja lärande, men lyfter även etiska och praktiska begränsningar. Ytterligare forskning krävs för att säkerställa långsiktiga läranderesultat och lärares möjligheter till bedömning inom spelifierade miljöer
Towards smarter grid forecasting : Predictive modeling of transformer load with AI
No full textTransformers that supply areas with electricity wear down over time as a result of fluctuating energy demands. This thesis focuses on the selection of the models and variables for deep learning and machine learning methods, particularly when used for electricity demand forecasts in transformers. By training and evaluating models on large datasets, we can compare the accuracy of the results. Our study examined what models were the most commonly top performing in the field of transformer load forecasting and subsequently compared Random Forest Regression, Extreme Gradient Boost, and Long Short-Term Memory at 50 different three day intervals and found that LSTM provided the most accurate results. The most influential variables in LSTM were air and sea temperature. Our findings show that there is potential for artificial intelligence to be used in transformer load forecasts
VR and 3D scanning for digital archiving : A conservational work on student-cultural places
No full textArbetet går ut på att undersöka hur en kulturell samlingsplats kan bevaras med hjälp av 3D-skanning, fotogrammetri och interaktiva VR-miljöer, samt att bevara en plats som har försvunnit under arbetets gång. De metoder som har använts för att skanna in platsen är: 3D-skanning med hjälp av en 3D-skanner, fotogrammetri med hjälp av en kamera och fotoskanning med skrivare. Det material som samlats in har sedan använts för att bygga upp en VR-miljö. Arbetet resulterade i en djupare förståelse kring styrkor och svagheter hos de olika metoderna samt strategier för att kringgå dem. För att få en verklighetstrogen gestaltning behövdes ett genomtänkt samspel mellan de olika metoderna vidtas och på så sätt utformades den interaktiva VR-miljön.This paper aims to explore how a cultural gathering place can be preserved using 3D scanning, photogrammetry, and interactive VR environments, as well as preserving a location that has disappeared during the course of the project. The methods used to scan the location are: 3D scanning with a 3D scanner, photogrammetry using a camera, and photo scanning with an printer. The material collected was then used to construct a VR environment. The project resulted in a deeper understanding of the strengths and weaknesses of the different methods, as well as strategies for overcoming them. In order to achieve a realistic representation, a well-considered interplay between the various methods was necessary, which ultimately shaped the interactive VR environment
Ergonomiskt knapp- och dragkedjehjälpmedel
No full textDetta projekt syftade till att förbättra ett hjälpmedel för att knäppa knappar och dra upp dragkedjor för personer med reumatiska sjukdomar, greppsvårigheter eller smärta och nedsatt rörlighet i fingrar och handleder. Befintliga produkter på marknaden upplevs ofta som svåra att använda, otillräckligt ergonomiska och exkluderar många inom målgruppen. Genom en användarcentrerad designprocess genomfördes förstudie, kravspecificering, konceptgenerering och utvärdering. Tre koncept togs fram, varav ett med vinklat handtag valdes vidare för utveckling då det underlättar rörelser med hela armen snarare än enbart handleden. Detaljkonstruktion, prototyptillverkning och användartester genomfördes, där det nya hjälpmedlet bedömdes mer bekvämt, intuitivt och visuellt tilltalande än ett kommersiellt alternativ. Projektet visar att även enkla ergonomiska förbättringar kan öka användbarheten och självständigheten. Arbetet utgör ett första steg mot en mer tillgänglig, värdig och funktionell design av vardagshjälpmedel.This project aimed to improve an assistive device for buttoning and zipping, designed for individuals with rheumatic diseases, grip difficulties, or pain and reduced mobility in fingers and wrists. Existing products on the market are often perceived as difficult to use, insufficiently ergonomic and exclude many within the target group. Through a user-centered design process, a preliminary study, requirement specification, concept generation and evaluation were conducted. Three concepts were developed, one of which, with an angled handle, was selected for further development, as it facilitates movement using the entire arm rather than just the wrist. Detailed design, prototype manufacturing and user testing were carried out. The new assistive device was rated as more comfortable, intuitive and visually appealing than a commercial alternative. This project shows that even simple ergonomic improvements can increase usability and independence. The work represents an initial step toward a more accessible, dignified and functional design of everyday assistive products
Market capitalization and shock resilience : European stock market responses to economic and non-economic surges
No full textThis study investigates how economic and non-economic shocks impact European stock market performance, with a particular focus on how firms of different market capitalizations, large, mid, and small, respond to systemic disruptions. Two pivotal events are analyzed: the Brexit referendum in 2016, representing a political shock, and the European energy crisis in 2022, representing an economic shock. An event study methodology is employed to measure abnormal returns and cumulative abnormal returns over both short-term and long-term event windows, allowing for a detailed assessment of the magnitude, timing, and persistence of market reactions. The results reveal that firm size significantly moderates stock price sensitivity to shocks, with small- and mid-cap firms displaying greater volatility and deeper abnormal returns compared to large-cap firms across both events. The Brexit referendum generated sharp, immediate market reactions, particularly among smaller firms more exposed to regulatory uncertainty and regional dependencies, with effects that persisted well beyond the initial shock. Conversely, the energy crisis led to a more gradual adjustment process, as the economic nature of the shock allowed investors time to reassess fundamentals, resulting in a slower deterioration of returns and partial recoveries over extended periods. A comparative analysis of these two events highlights important temporal and structural differences: political shocks tend to trigger concentrated, front-loaded declines driven by ambiguity and investor sentiment, while economic shocks disseminate more progressively through financial systems, affecting corporate performance via operational and macroeconomic channels. These findings align with historical evidence from prior crises, such as the Global Financial Crisis and the COVID-19 pandemic, reinforcing the critical importance of firm size, financial flexibility, and international exposure in determining resilience to external shocks. However, it should be noted that several statistical tests, particularly non-parametric and cross-sectional methods, did not confirm significance across all indices and event windows. This highlights some limitations in statistical robustness, especially in long-term estimations and during periods of low volatility. This research offers valuable contributions to the literature by providing a comparative perspective on political versus economic event impacts, emphasizing the heterogeneous effects across market capitalizations and time horizons. Practical implications are drawn for investors aiming to optimize risk management strategies and for policymakers seeking to enhance financial system stability. Limitations regarding event selection, regional scope, and methodological assumptions are acknowledged, suggesting avenues for future research, including cross-regional comparative studies, sector-specific analyses, and exploration of policy intervention effects in moderating shock transmission