3D reconstruction can be used in forensic science to reconstruct crime scenes and objects so that measurements and further information can be acquired off-site. It is desirable to use image based reconstruction methods but there is currently no procedure available for determining the uncertainty of such reconstructions. In this thesis the uncertainty of Structure from Motion is investigated. This is done by exploring the literature available on the subject and compiling the relevant information in a literary summary. Also, Monte Carlo simulations are conducted to study how the feature position uncertainty affects the uncertainty of the parameters estimated by bundle adjustment. The experimental results show that poses of cameras that contain few image correspondences are estimated with higher uncertainty. The poses of such cameras are estimated with lesser uncertainty if they have feature correspondences in cameras that contain a higher number of projections.

The problem of determining who should do what given a set of tasks and a set of agents is called the task allocation problem. The problem occurs in many multi-agent system applications where a workload of tasks should be shared by a number of agents. In our case, the task allocation problem occurs as an integral part of a larger problem of determining if a task can be delegated from one agent to another. Delegation is the act of handing over the responsibility for something to someone. Previously, a theory for delegation including a delegation speech act has been specified. The speech act specifies the preconditions that must be fulfilled before the delegation can be carried out, and the postconditions that will be true afterward. To actually use the speech act in a multi-agent system, there must be a practical way of determining if the preconditions are true. This can be done by a process that includes solving a complex task allocation problem by the agents involved in the delegation. In this thesis a constraint-based task specification formalism, a complex task allocation algorithm for allocating tasks to unmanned aerial vehicles and a generic collaborative system shell for robotic systems are developed. The three components are used as the basis for a collaborative unmanned aircraft system that uses delegation for distributing and coordinating the agents' execution of complex tasks.

Being able to train machine learning models on simulated data can be of great interest in several applications, one of them being for autonomous driving of cars. The reason is that it is easier to collect large labeled datasets as well as performing reinforcement learning in simulations. However, transferring these learned models to the real-world environment can be hard due to differences between the simulation and the reality; for example, differences in material, textures, lighting and content. One approach is to use domain adaptation, by making the simulations as similar as possible to the reality. The thesis's main focus is to investigate domain adaptation as a way to meet the reality-gap, and also compare it to an alternative method, domain randomization. Two different methods of domain adaptation; one adapting the simulated data to reality, and the other adapting the test data to simulation, are compared to using domain randomization. These are evaluated with a classifier making decisions for a robot car while driving in reality. The evaluation consists of a quantitative evaluation on real-world data and a qualitative evaluation aiming to observe how well the robot is driving and avoiding obstacles. The results show that the reality-gap is very large and that the examined methods reduce it, with the two using domain adaptation resulting in the largest decrease. However, none of them led to satisfactory driving.

What people take for granted is not as easy for computers. Being able tojudge whether an image is the same even though it has a differentresolution or is taken from a different angle or light condition is easyfor humans but much more difficult for computers. Today’s mobiles aremore powerful than ever, which has opened up for more hardware-demandingalgorithms to be processed. How to effectively match handwritten notesto eliminate duplicates in an application. Are there better or worsemethods and approaches, and how do they compare to each other? Can youachieve both accuracy and speed? By analyzing images taken at differentangles, distances, and lighting conditions, different methods andapproaches have been developed and analyzed. The methods are representedin various tables where time and accuracy are represented. Eightdifferent methods were evaluated. The methods were tuned on one datasetconsisting of 150 post-it notes, each imaged under four conditions,leading to 600 images and 1800 possible pair-wise matches. The methodswere thereafter evaluated on an independent dataset consisting of 250post-it notes, each imaged under four conditions, leading to 1000 imagesand 3000 possible pair-wise matches. The best method found 99.7%, andthe worst method found 62.9% of the matching pairs. Seven of the eightevaluated matches did not make any incorrect matches. / Det människor tar för givet är inte lika lätt för datorer. Att kunna bedöma om en bild är den samma fast den har annan upplösning eller är tagen från en annan vinkel eller ljusförhållande är lätt för människor men betydligt svårare för datorer. Dagens mobiler är kraftfullare än någonsin vilket har öppnat upp för att mer hårdvaru krävande algoritmer kan processas. Hur matchar man handskrivna lappar på ett effektivt sätt för att eliminera kopior i en applikation. Finns det bättre eller sämre metoder och tillvägagångssätt, och hur står de gentemot varandra? Kan man uppnå både träffsäkerhet samt snabbhet? Genom att analysera bilder tagna får olika vinklar, avstånd samt ljusförhållanden har olika metoder och tillvägagångssätt utvecklats och analyserats. Metoderna är representerade i olika tabeller där tid, samt träffsäkerhet redovisas. Åtta olika metoder utvärderades. Metoderna ställdes in på ett dataset bestående av 150 post-it-lappar, var och en fotograferade under fyra förhållanden, vilket ledde till 600 bilder och 1800 möjliga matchningar. Metoderna utvärderades därefter på ett oberoende dataset bestående av 250 post-it-lappar, var och en fotograferade under fyra förhållanden, vilket ledde till 1000 bilder och 3000 möjliga matchningar. Den bästa metoden fann 99,7% och den sämsta metoden hittade 62,9% av de matchande paren. Sju av de åtta utvärderade metoderna gjorde inga felaktiga matchningar.

Images are often corrupted by noise which reduces their visual quality and interferes with analysis. Convolutional Neural Networks (CNNs) have become a popular method for denoising images, but their training typically relies on access to thousands of pairs of noisy and clean versions of the same underlying picture. Unsupervised methods lack this requirement and can instead be trained purely using noisy images. This thesis evaluated two different unsupervised denoising algorithms: Noise2Self (N2S) and Parametric Probabilistic Noise2Void (PPN2V), both of which train an internal CNN to denoise images. Four different CNNs were tested in order to investigate how the performance of these algorithms would be affected by different network architectures. The testing used two different datasets: one containing clean images corrupted by synthetic noise, and one containing images damaged by real noise originating from the camera used to capture them. Two of the networks, UNet and a CBAM-augmented UNet resulted in high performance competitive with the strong classical denoisers BM3D and NLM. The other two networks - GRDN and MultiResUNet - on the other hand generally caused poor performance.

As the Automotive industry is heavily regulated from a quality point of view, excellence in pro-duction is obligatory. Due to the fact that removing human error from humans is impossible, new solutions must be found. The transition to more data driven production strategies enables the implantation of automated vision systems for replacing humans in simple classification tasks. As research in the field of artificial intelligence advances, the hardware required to run the algorithms decreases. Concurrently small computing platforms break new performance records and the innovation space converges. This work harnesses state-of-the-art from both domains by implementing a plug-on vision system, driven by a resource-constrained edge device in a production line. The implemented CNN-model based on the MobileNetV2 architecture achieved 97.80, 99.93, and 95.67% in accuracy, precision, and recall respectively. The model was trained using only 100 physical samples, which were expanded by a ratio of 1:15 through innovative real world and digital augmentations. The core of the vision system was a commodity device, the Raspberry Pi 4. The solution fulfilled all the requirements while sparking new development ideas for future work.

In the digital age where video content is abundant, this thesis investigates the efficient adaptation of an existing video-language model (VLM) to new data. The research leverages CLIP, a robust language-vision model, for various video-related tasks including video retrieval. The study explores using pre-trained VLMs to extract video embeddings without the need for extensive retraining. The effectiveness of a smaller model using aggregation is compared with larger models and the application of logistic regression for few-shot learning on video embeddings is examined. The aggregation was done using both non-learning through mean-pooling and also by utilizing a transformer. The video-retrieval models were evaluated on the ActivityNet Captions dataset which contains long videos with dense descriptions while the linear probes were evaluated on ActivityNet200 a video classification dataset.  The study's findings suggest that most models improved when additional frames were employed through aggregation, leading to improved performance. A model trained with fewer frames was able to surpass those trained with two or four times more frames by instead using aggregation. The incorporation of patch dropout and the freezing of embeddings proved advantageous by enhancing performance and conserving training resources. Furthermore, using a linear probe showed that the extracted features were of high quality requiring only 2-4 samples per class to match the zero-shot performance.

Denna kandidatuppsats utforskar framtiden för jailbreaking i det svenska samhället och de risker som är förknippade med att utföra jailbreak på personbilar. Rapporten presenterar välförankrade scenarion för ämnets framtid och hur det kan existera inom olika scenarier. Rapporten baseras på intervjuer med olika aktörer inom hela branschen, såväl som den reglerande myndigheten och en konsumentorganisation. Tillsammans med en litteraturstudie av sårbarheterna i personbilars mjukvaror och dess användning, samt en mer ingående undersökning av jailbreaking-gemenskapens historia, används "Six pillars approach" för att göra förutsägelser om framtiden för jailbreaking och hur det kan få en plats i den framtida debatten. Genom att dra paralleller till dagens smartphones visar resultaten att det finns ett antal scenarier som kommer att påverka hur vanligt förekommande jailbreaking kommer att vara, men det är troligt att det kommer att existera i någon skepnad. / This thesis explores the future of jailbreaking in Swedish society and the risks associated with performing a jailbreak on a vehicle. It attempts to paint well founded visualizations of the future of this topic and how it may exist within different scenarios. It is based on the interviews with different actors and stakeholders within the entire industry as well as the regulatory body and a consumer organization. Combined with a literature study of the vulnerabilities present in vehicle software and its usage, in addition to a deeper dive into the history of the jailbreaking community, it uses the “Six pillars approach” to make predictions for the future of jailbreaking and how it may have a place in the future debate. Drawing some parallels to today's smartphones, the results show that there are a number of scenarios that will impact how common jailbreaking will be, but that it is somewhat likely that it will existin some capacity.

Idag är det vanligt att det sker filmning av evenemang utan att man använder sig av en professionell videofotograf. Det kan vara knatteligans fotbollsmatch, konferensmöten, undervisning eller YouTube-klipp. För att slippa ha en kameraman kan man använda sig av något som kallas för objektföljningskameror. Det är en kamera som kan följa ett objekts position över tid utan att en kameraman styr kameran. I detta examensarbete beskrivs hur objektföljning fungerar, samt görs en jämförelse mellan objektföljningskameror med datorseende och en kameraman. För att kunna jämföra de mot varandra har en prototyp byggts. Prototypen består av en Raspberry Pi 4B med MOSSE som är en objektföljningsalgoritm och SSD300 som är en detekteringsalgoritm inom datorseende. Styrningen består av en gimbal som består av tre borstlösa motorer som styr kameran med en regulator. Resultatet blev en prototyp som klarar av att följa en person som promenerar i maximalt 100 pixlar per sekund eller 1 meter per sekund i helbild, med en maxdistans på 11,4 meter utomhus. Medan en kameraman klarar av att följa en person i 300–800 pixlar per sekund eller 3 meter per sekund. Prototypen är inte lika bra som en ka-meraman men kan användas för att följa en person som undervisar och går långsamt. Under förutsättningen att prototypen är robust vilket inte är fallet. För att få bättre resultat behövs starkare processor och bättre algoritmer än som använts med prototypen. Då ett stort problem var att uppdateringshastigheten var låg för detekteringsalgoritmen. / Today, it is common for events to be filmed without the use of a professional video photographer. It can be the little league football game, conference meetings, teaching or YouTube clips. To film without a cameraman, you can use something called object tracking cameras. It is a camera that can follow an object's position without a cameraman.This thesis describes how object tracking works as well as comparison between ob-ject tracking cameras with computer vision and a cameraman. In order to compare them against each other, a prototype has been developed. The prototype consists of a Raspberry Pi 4B with MOSSE which is an object tracking algorithm and SSD300 which is a detection algorithm in computer vision. The steering consists of a gimbal consisting of three brushless motors that control the camera with a regulator. The result was a prototype capable of following a person walking at a maximum speed 100 pixels per second or 1 meter per second in full screen, with a maximum distance of 11.4 meters outdoors. While a cameraman managed to follow a person at 300-800 pixels per second or 3 meters per second. The prototype is not as good as a cameraman but can be used to follow a person who teaches and walks slowly. Under basis that the prototype is robust, which is not the case. To get better results, stronger processor and better algorithms are needed than used with the prototype. That’s because a big problem was that the refresh rate was low for the detection algorithm.

Human pose estimation in 3D is a large area within computer vision, with many application areas. A common approach is to first estimate the pose in 2D, resulting in a confidence heatmap, and then estimate the 3D pose using the most likely estimations in 2D. This may, however, cause problems in cases where pose estimates are more uncertain and the estimation of one point is far from the true position, for example when a limb is occluded. This thesis adapts the method Neural Radiance Fields (NeRF) to 2D confidence heatmaps in order to create an implicit representation of the uncertainty in 3D, thus attempting to make use of as much information in 2D as possible. The adapted method was evaluated on the Human3.6M dataset, and results show that this method outperforms a simple triangulation baseline, especially when the estimation in 2D is far from the true pose.