The introduction of autonomous vehicles in traffic is driven by expected gains in multiple areas, such as improvement of health and safety, better resource utilization, pollution reduction and greater convenience. The development of more competent algorithms will determine the rate and level of success for the ambitions around autonomous vehicles. In this thesis work an intersection management system for a mix of autonomous and manually driven vehicles is created. The purpose is to investigate the strategy to combine turn intention prediction for manually driven vehicles with scheduling of autonomous vehicle. The prediction method used is support vector machine (SVM) and scheduling of vehicles have been made by dividing the intersection into an occupancy grid and apply different safety levels. Real-life data comprising recordings of large volumes of traffic through an intersection has been combined with simulated vehicles to assess the relevance of the new algorithms. Measurements of collision rate and traffic flow showed that the algorithms behaved as expected. A miniature vehicle based on a prototype for an autonomous RC-car has been designed with the purpose of testing of the algorithms in a laboratory setting. / Införandet av autonoma fordon i trafiken drivs av förväntade vinster i flera områden, såsom förbättring av hälsa och säkerhet, bättre resursutnyttjande, minskning av föroreningar och ökad bekvämlighet. Utvecklingen av mer kompetenta algoritmer kommer att bestämma hastigheten och nivån på framgång för ambitionerna kring autonoma fordon. I detta examensarbete skapas ett korsningshanteringssystem för en blandning av autonoma och självkörande bilar. Syftet är att undersöka strategin att kombinera prediktion av hur manuellt styrda bilar kommer att svänga med att schemalägga autonoma bilar utifrån detta. Prediktionsmetoden som använts är support vector machine (SVM) och schemaläggning av bilar har gjorts genom att dela upp korsningen i ett occupancy grid och tillämpa olika säkerhetsmarginaler. Verklig data från inspelningar av stora volymer trafik genom en korsning har kombinerats med simulerade fordon för att bedöma relevansen av de nya algoritmerna. Mätningar av kollisioner och trafikflöde visade att algoritmerna uppträdde som förväntat. Ett miniatyrfordon baserat på en prototyp av en självkörande radiostyrd bil har tagits fram i syfte att testa algoritmerna i laboratoriemiljö.

The transport industry is experiencing a disruption as fully autonomous vehicles are introduced in traffic. The intelligent, driverless vehicles will reduce cost, liberate human effort and increase safety. Today, the hardware technology seems to have reached the required processing power, but the decision-making algorithm still has a long way to go until they’re proven to be road-safe. Among these is the problem of lateral path tracking control. This thesis will consider the lateral control problem with the goal to send the right signal to the steering actuators so that the vehicle follows a predetermined trajectory. The vehicle in question is a triaxial, rigid, electric truck with active steering on both front and rearmost wheels. With servo latency and large inertial parameters in mind, a highly accurate model of the lateral and yaw behavior must be identified in order to predict the vehicle dynamics for a given steering input. Then, the properties of an optimal lateral controller are iteratively improved until a sufficiently low tracking error is obtained. Lastly, the controller is tuned to guarantee robustness for a range of uncertain vehicle parameters. The derived triaxial model with servo actuation is proven to be better at predicting the vehicle dynamics compared to other models common in literature with only one active steering input. When constructing a lateral controller, the importance was shown of considering 1) state feedback control of the lateral error, 2) feedforward control operating on future road curvature, 3) integrating control which combats biases and model errors, 4) using a tailored triaxial model and 5) minimizing the control signal change. Lastly, the derived controller was shown to have a decent stability margin with respect to estimated uncertainties. / Transportbranschen är i ett skifte då helt autonoma fordon införs i trafiken. De intelligenta, förarlösa fordonen minskar kostnader, ökar säkerheten och låter oss människor syssla med annat. Idag verkar det som att hårdvarutekniken har den processorkraft som behövs men de beslutsfattande algoritmerna har fortfarande en lång väg att gå tills de har visat sig vara helt vägsäkra. Bland dessa är problemet med lateral styrningskontroll som kommer ses över i denna avhandling. Fordonet i fråga är en rigid lastbil med tre hjulaxlar och aktiv styrning på både de främre och bakersta hjulen. Med tanke på servofördröjningar och de stora tröghetsparametrarna måste en noggrann modell av dynamiken identifieras för att förutspå responsen för en viss styrvinkel. Därefter utvecklas en optimal lågnivåregulator iterativt tills ett tillräckligt lågt spårningsfel erhållits. Slutligen ställs regulatorn in för att garantera robusthet för ett set av osäkra fordonsparametrar Den härledda triaxialmodellen med servostyrning var bevisbart bättre på att förutspå fordonsdynamiken jämfört med andra modeller som återkommer frekvent i litteraturen. Vid regulatorkonstruktionen påvisades vikten av att överväga 1) återkoppling av laterala felet, 2) förhandsgranskning som tittar på den kommande vägkrökningen, 3) integrering av styrfelet som åtgärdar modellfel, 4) en skräddarsydd fordonsmodell med tre axlar och 5) minimering av ändringen utav kontrollsignalen. Slutligen visades den härledda regulatorn ha en skaplig stabilitetsmarginal gentemot uppskattade osäkerheter av parametrar.

The purpose of this study is to investigate what underlying causes drive people to take part in the Swedish autonomous left-wing movement. The autonomous left-wing movement is an extraparliamentary mob, in Sweden often attributed in negative terms, considered as extremists. Interviewing six activists has given me the chance to better understand the movement in terms of ideology as well as understanding their lack of reliability in the Swedish government and other social structures. The study also shows a complexity in the relationship between the interviewees and fascists, as well as in the Swedish police and the Swedish migration office, which is based on a combination of ideological values as well as a lack of trust to the Swedish system as a whole.

In recent years semantic segmentation models utilizing Convolutional Neural Networks (CNN) have seen significant success for multiple different segmentation problems. Models such as U-Net have produced promising results within the medical field for both regular 2D and volumetric imaging, rivalling some of the best classical segmentation methods. In this thesis we examined the possibility of using a convolutional neural network-based model to perform segmentation of discrete bone fragments in CT-volumes with segmentation-hints provided by a user. We additionally examined different classical segmentation methods used in a post-processing refinement stage and their effect on the segmentation quality. We compared the performance of our model to similar approaches and provided insight into how the interactive aspect of the model affected the quality of the result. We found that the combined approach of interactive segmentation and deep learning produced results on par with some of the best methods presented, provided there were adequate amount of annotated training data. We additionally found that the number of segmentation hints provided to the model by the user significantly affected the quality of the result, with convergence of the result around 8 provided hints.

In large scale productions of metal sheets, it is important to maintain an effective way to continuously inspect the products passing through the production line. The inspection mainly consists of detection of defects and tracking of ID numbers. This thesis investigates the possibilities to create an automatic inspection system by evaluating different machine learning algorithms for defect detection and optical character recognition (OCR) on metal sheet data. Digit recognition and defect detection are solved separately, where the former compares the object detection algorithm Faster R-CNN and the classical machine learning algorithm NCGF, and the latter is based on unsupervised learning using a convolutional autoencoder (CAE). The advantage of the feature extraction method is that it only needs a couple of samples to be able to classify new digits, which is desirable in this case due to the lack of training data. Faster R-CNN, on the other hand, needs much more training data to solve the same problem. NCGF does however fail to classify noisy images and images of metal sheets containing an alloy, while Faster R-CNN seems to be a more promising solution with a final mean average precision of 98.59%. The CAE approach for defect detection showed promising result. The algorithm learned how to only reconstruct images without defects, resulting in reconstruction errors whenever a defect appears. The errors are initially classified using a basic thresholding approach, resulting in a 98.9% accuracy. However, this classifier requires supervised learning, which is why the clustering algorithm Gaussian mixture model (GMM) is investigated as well. The result shows that it should be possible to use GMM, but that it requires a lot of GPU resources to use it in an end-to-end solution with a CAE.

Semantic segmentation is a key approach to comprehensive image data analysis. It can be applied to analyze 2D images, videos, and even point clouds that contain 3D data points. On the first two problems, CNNs have achieved remarkable progress, but on point cloud segmentation, the results are less satisfactory due to challenges such as limited memory resource and difficulties in 3D point annotation. One of the research studies carried out by the Computer Vision Lab at Linköping University was aiming to ease the semantic segmentation of 3D point cloud. The idea is that by first projecting 3D data points to 2D space and then focusing only on the analysis of 2D images, we can reduce the overall workload for the segmentation process as well as exploit the existing well-developed 2D semantic segmentation techniques. In order to improve the performance of CNNs for 2D semantic segmentation, the study has used input data derived from different modalities. However, how different modalities can be optimally fused is still an open question. Based on the above-mentioned study, this thesis aims to improve the multistream framework architecture. More concretely, we investigate how different singlestream architectures impact the multistream framework with a given fusion method, and how different fusion methods contribute to the overall performance of a given multistream framework. As a result, our proposed fusion architecture outperformed all the investigated traditional fusion methods. Along with the best singlestream candidate and few additional training techniques, our final proposed multistream framework obtained a relative gain of 7.3\% mIoU compared to the baseline on the semantic3D point cloud test set, increasing the ranking from 12th to 5th position on the benchmark leaderboard.

Syftet med denna studie var att undersöka det framtida användandet av instrumentpanelen i autonoma personbilar. En lösning presenteras där interiören liknar en tågkupé med säten vända mot varandra kring den nya instrumentpanelen utformad som ett multifunktionellt bord. Genom att undersöka vad konsumenter från fyra olika målgrupper ville sysselsätta sig med i en nivå fem autonom personbil kunde funktioner såsom bildskärmar, tangentbord och förvaringsmöjligheter inkluderas i den nya instrumentpanelen.   För insamling av empiri användes både kvalitativa och kvantitativa metoder där semistrukturerade intervjuer och en enkätundersökning genomfördes. Båda metoderna riktade sig till fyra målgrupper av konsumenter: Studenter, barnfamiljer, kortvägspendlare samt resande säljare. För att samla in mycket information på kort tid utfördes metoderna samtidigt och båda metoderna användes för att validera resultatet.   Genom analysen av empirin hittades gemensamma intressen mellan målgrupperna, i båda metoderna, och sammanställde dessa till kundönskemål. Resultatet av analysen visar bland annat att passagerare i autonoma fordon vill ha bra möjligheter till att arbeta, lyssna på musik, docka telefon, laptop eller surfplatta till inbyggda skärmar i bilen samt läsa och skriva email. Det önskas hållare för drycker, avlastningsytor för mat samt kyld förvaring.   Intervjuer med experter från branschen genomfördes för att bistå med utformning- och säkerhetskrav som tillsammans med kundönskemålen gav ett underlag för att generera koncept. Innan konceptegenereringsfasen påbörjades gjordes en brainstorming för att diskutera tekniska lösningar till de framtagna önskemålen. De framtagna koncepten utvärderades med metoden för Pughs konceptvalsmatris där de mättes mot ett referenskoncept. Ett vinnande koncept kunde efter förbättringar utses och presenteras med skisser, produktbeskrivning samt en produktspecifikation.  Arbetet begränsades till att fokusera på att uppfylla kundönskemålen och lämnar många krav runt säkerhet åt framtida vidareutveckling av konceptet. / Contents of this bachelor’s thesis are written in Swedish.  The purpose of this study was to investigate the future use of the instrument panel in autonomous cars. A solution is presented in which the interior resembles a train compartment with seats facing each other around the new instrument panel designed as a multifunctional table. By examining what consumers from four different target groups would want to engage themselves with in a level five autonomous car, features such as monitors, keyboards and storage facilities was included in the new instrument panel.  For the gathering of empirical data, qualitative and quantitative methods was used, where both semi-structured interviews and a survey was conducted. Both methods addressed four target groups of consumers: students, families with children, short-distance commuters and traveling salespersons. To collect much information in a short period of time, the methods were performed simultaneously and both methods were used to validate the result.  The empirical analysis found common interests between the target groups, in both methods and compiled these into customer requests. The result of the analysis shows, among other things, that passengers in autonomous cars want good opportunities to work, listen to music, dock their phone, laptop or tablet too built-in monitors in the car as well as read and write email. They desired holder for drinks, relief surfaces when eating food as well as refrigerated storage.  Interviews with industry experts were conducted to complement with design and safety requirements that, together with customer requests, provided a basis for generating concepts. Before the start of the concept generating phase, a brainstorming was conducted to discuss technical solutions to the desired customer requests. The final concepts were evaluated using the method of Pugh Concept Selection, where they were compared against a reference concept. A winning concept was, after improvements, presented with sketches, product description and a product specification.  The work was limited to focusing on meeting customer requests and leaving many requirements for personal safety to future, further development of the concept.

In this thesis, three different neural network architectures are investigated to detect the action of a shot within a football game using video data. The first architecture uses con- ventional convolution and pooling layers as feature extraction. It acts as a baseline and gives insight into the challenges faced during shot detection. The second architecture uses a pre-trained feature extractor. The last architecture uses three-dimensional convolution. All these networks are trained using short video clips extracted from football game video streams. Apart from investigating network architectures, different sampling methods are evaluated as well. This thesis shows that amongst the three evaluated methods, the ap- proach using MobileNetV2 as a feature extractor works best. However, when applying the networks to a video stream there are a multitude of challenges, such as false positives and incorrect annotations that inhibit the potential of detecting shots.

Background: Video surveillance is a growing area where it can help with deterring crime, support investigation or to help gather statistics. These are just some areas where video surveillance can aid society. However, there is an improvement that could increase the efficiency of video surveillance by introducing tracking. More specifically, tracking between cameras in a network. Automating this process could reduce the need for humans to monitor and review since the tracking can track and inform the relevant people on its own. This has a wide array of usability areas, such as forensic investigation, crime alerting, or tracking down people who have disappeared. Objectives: What we want to investigate is the common setup of real-time multi-target multi-camera tracking (MTMCT) systems. Next up, we want to investigate how the components in an MTMCT system affect each other and the complete system. Lastly, we want to see how image enhancement can affect the MTMCT. Methods: To achieve our objectives, we have conducted a systematic literature review to gather information. Using the information, we implemented an MTMCT system where we evaluated the components to see how they interact in the complete system. Lastly, we implemented two image enhancement techniques to see how they affect the MTMCT. Results: As we have discovered, most often, MTMCT is constructed using a detection for discovering object, tracking to keep track of the objects in a single camera and a re-identification method to ensure that objects across cameras have the same ID. The different components have quite a considerable effect on each other where they can sabotage and improve each other. An example could be that the quality of the bounding boxes affect the data which re-identification can extract. We discovered that the image enhancement we used did not introduce any significant improvement. Conclusions: The most common structure for MTMCT are detection, tracking and re-identification. From our finding, we can see that all the component affect each other, but re-identification is the one that is mostly affected by the other components and the image enhancement. The two tested image enhancement techniques could not introduce enough improvement, but other image enhancement could be used to make the MTMCT perform better. The MTMCT system we constructed did not manage to reach real-time.

The concept of Computer Vision is not new or fresh. On contrary ideas have been shared and worked on for almost 60 years. Many use cases have been found throughout the years and various systems developed, but there is always a place for improvement. An observation was made, that methods used today are generally focused on a single purpose and implement expensive technology, which could be improved. In this report, we are going to go through an extensive research to find out if a professionally sold, expensive software, can be replaced by an off the shelf, low-cost solution entirely designed and developed in-house. To do that we are going to look at the history of Computer Vision, examples of applications, algorithms, and find general scenarios or computer vision problems which can be solved. We are then going take a step further and define solid use cases for each of the scenarios found. Finally, a prototype solution is going to be designed and presented. After analysing the results gathered we are going to reach out to the reader convincing him/her that such application can be developed and work efficiently in various areas saving investments to businesses.