Gaze tracking means to detect and follow the direction in which a person looks. This can be used in for instance human-computer interaction. Most existing systems illuminate the eye with IR-light, possibly damaging the eye. The motivation of this thesis is to develop a truly non-intrusive gaze tracking system, using only a digital camera, e.g. a web camera. The approach is to detect and track different facial features, using varying image analysis techniques. These features will serve as inputs to a neural net, which will be trained with a set of predetermined gaze tracking series. The output is coordinates on the screen. The evaluation is done with a measure of accuracy and the result is an average angular deviation of two to four degrees, depending on the quality of the image sequence. To get better and more robust results, a higher image quality from the digital camera is needed.

Fluoroskopi är benämningen på kontinuerlig röntgengenomlysning av en patient. Eftersom patienten och även läkaren då utsätts för kontinuerlig röntgenstrålning måste strålningsdosen hållas låg, vilket leder till brusiga bilder. Det är därför önskvärt att genom bildbehandling förbättra bilderna. Bildförbättringen måste dock ske i realtid och därför kan inte konventionella metoder användas. Detta examensarbete avser att undersöka hur ortogonala s k. derivataoperatorer kan användas för att förbättra läsbarheten av fluoroskopibilder med hjälp av brusundertryckning och kantförstärkning. Derivataoperatorer är separerbara vilket gör dem extremt beräkningsvänliga och lätta att infoga i en skalpyramid. Skalpyramiden ger möjlighet att processa strukturer och detaljer av olika storlek var för sig samtidigt som nedsamplingsmekanismen gör att denna uppdelning inte nämnvärt ökar beräkningsbördan. I den fullständiga lösningen införes också struktur-/brusseparering för att förhindra förstärkning av och undertrycka bidrag från de frekvensband där en pixel domineras av brus. Resultaten visar att brus verkligen kan undertryckas medan kanter och linjer bevaras bra eller förstärkes om så önskas. Den riktade filtreringen gör dock att det lätt uppstår maskliknande strukturer i bruset, men detta kan undvikas med rätt parameterinställning av struktur-/brussepareringen. Förhållandet mellan riktad och icke-riktad filtrering är likaledes styrbart via en parameter som kan optimeras med hänsyn till behov och önskemål vid varje tillämpning. / In X-ray technology, fluoroscopy stands for continuous irradiation. For the sake of both patients and doctors the dose has to be kept low, which leads to noisy images and the question of possible enhancement by digital image processing. Since such enhancement has to be done in real-time, most conventional and available methods are unsuitable. The purpose of this thesis is to examine how derivative operators can be used to improve fluoroscopy images in terms of noise reduction and edge enhancement. Since the derivative operators are designed as highly separable convolution kernels the image derivatives can be computed very efficiently with a scheme that is readily embedded in a scale-space pyramid. In this pyramid, structures and details of different sizes can be processed separately with optimal parameter settings. In the final solution we also discriminate between structure and noise in order to avoid amplification, even suppress contributions from frequency bands where a certain pixel position is dominated by noise. Experimental results show that noise can indeed be suppressed while edges and lines are enhanced. Oriented filtering may induce false structures in areas where only noise is present, something that can be avoided by correcting the parameters in the noise/structure discriminator. The relation between oriented and non-oriented filtering is likewise controllable with a parameter that can be optimized for application dependent needs and desires.

This thesis presents a complete and fully automatic method for estimating the volume of an airbag, through all stages of its inflation, with multiple synchronized high-speed cameras. Using recorded contours of the inflating airbag, its visual hull is reconstructed with a novel method: The intersections of all back-projected contours are first identified with an accelerated epipolar algorithm. These intersections, together with additional points sampled from concave surface regions of the visual hull, are then Delaunay triangulated to a connected set of tetrahedra. Finally, the visual hull is extracted by carving away the tetrahedra that are classified as inconsistent with the contours, according to a voting procedure. The volume of an airbag's visual hull is always larger than the airbag's real volume. By projecting a known synthetic model of the airbag into the cameras, this volume offset is computed, and an accurate estimate of the real airbag volume is extracted. Even though volume estimates can be computed for all camera setups, the cameras should be specially posed to achieve optimal results. Such poses are uniquely found for different airbag models with a separate, fully automatic, simulated annealing algorithm. Satisfying results are presented for both synthetic and real-world data.

Visual tracking is a classical computer vision problem with many important applications in areas such as robotics, surveillance and driver assistance. The task is to follow a target in an image sequence. The target can be any object of interest, for example a human, a car or a football. Humans perform accurate visual tracking with little effort, while it remains a difficult computer vision problem. It imposes major challenges, such as appearance changes, occlusions and background clutter. Visual tracking is thus an open research topic, but significant progress has been made in the last few years. The first part of this thesis explores generic tracking, where nothing is known about the target except for its initial location in the sequence. A specific family of generic trackers that exploit the FFT for faster tracking-by-detection is studied. Among these, the CSK tracker have recently shown obtain competitive performance at extraordinary low computational costs. Three contributions are made to this type of trackers. Firstly, a new method for learning the target appearance is proposed and shown to outperform the original method. Secondly, different color descriptors are investigated for the tracking purpose. Evaluations show that the best descriptor greatly improves the tracking performance. Thirdly, an adaptive dimensionality reduction technique is proposed, which adaptively chooses the most important feature combinations to use. This technique significantly reduces the computational cost of the tracking task. Extensive evaluations show that the proposed tracker outperform state-of-the-art methods in literature, while operating at several times higher frame rate. In the second part of this thesis, the proposed generic tracking method is applied to human tracking in surveillance applications. A causal framework is constructed, that automatically detects and tracks humans in the scene. The system fuses information from generic tracking and state-of-the-art object detection in a Bayesian filtering framework. In addition, the system incorporates the identification and tracking of specific human parts to achieve better robustness and performance. Tracking results are demonstrated on a real-world benchmark sequence.

Brown adipose tissue (BAT) is the main tissue involved in non-shivering heat production. A greater understanding of BAT could possibly lead to new ways of prevention and treatment of obesity and type 2 diabetes. The increasing prevalence of these conditions and the problems they cause society and individuals make the study of the subject important. An ongoing study performed at the Turku University Hospital uses images acquired using PET/MRI with 18F-FDG as the tracer. Scans are performed on sedentary and athlete subjects during normal room temperature and during cold stimulation. Sedentary subjects then undergo scanning during cold stimulation again after a six weeks long exercise training intervention. This degree project used images from this study. The objective of this degree project was to examine methods to automatically and objectively quantify parameters relevant for activation of BAT in combined PET/MRI data. A secondary goal was to create images showing glucose uptake changes in subjects from images taken at different times. Parameters were quantified in adipose tissue directly without registration (image matching), and for neck scans also after registration. Results for the first three subjects who have completed the study are presented. Larger registration errors were encountered near moving organs and in regions with less information. The creation of images showing changes in glucose uptake seem to be working well for the neck scans, and somewhat well for other sub-volumes. These images can be useful for identification of BAT. Examples of these images are shown in the report.

Image based modeling using visual images has undergone major development during the earlier parts of the 21th century. In this thesis a system for automated uncalibrated scene reconstruction using infrared images is implemented and tested. An automated reconstruction system could serve to simplify thermal inspection or as a demonstration tool. Thermal images will in general have lower resolution, less contrast and less high frequency content as compared to visual images. These characteristics of infrared images further complicates feature extraction and matching, key steps in the reconstruction process. In order to remedy the complication preprocessing methods are suggested and tested as well. Infrared modeling will also impose additional demands on the reconstruction as it is of importance to maintain thermal accuracy of the images in the product. Three main results are obtained from this thesis. Firstly, it is possible to obtain camera calibration and pose as well as a sparse point cloud reconstruction from an infrared image sequence using the suggested implementation. Secondly, correlation of thermal measurements from the images used to reconstruct three dimensional coordinates is presented and analyzed. Lastly, from the preprocessing evaluation it is concluded that the tested methods are not suitable. The methods will increase computational cost while improvements in the model are not proportional. / Bildbaserad modellering med visuella bilder har genomgått en stor utveckling under de tidigare delarna av 2000-talet. Givet en sekvens bestående av vanliga tvådimensionella bilder på en scen från olika perspektiv så är målet att rekonstruera en tredimensionell modell. I denna avhandling implementeras och testas ett system för automatiserad okalibrerad scenrekonstruktion från infraröda bilder. Okalibrerad rekonstruktion refererar till det faktum att parametrar för kameran, såsom fokallängd och fokus, är okända och enbart bilder används som indata till systemet. Ett stort användingsområde för värmekameror är inspektion. Temperaturskillnader i en bild kan indikera till exempel dålig isolering eller hög friktion. Om ett automatiserat system kan skapa en tredimensionell modell av en scen så kan det bidra till att förenkla inspektion samt till att ge en bättre överblick. Värmebilder kommer generellt att ha lägre upplösning, mindre kontrast och mindre högfrekvensinnehåll jämfört med visuella bilder. Dessa egenskaper hos infraröda bilder komplicerar extraktion och matchning av punkter i bilderna vilket är viktiga steg i rekonstruktionen. För att åtgärda komplikationen förbehandlas bilderna innan rekonstruktionen, ett urval av metoder för förbehandling har testats. Rekonstruktion med värmebilder kommer också att ställa ytterligare krav på rekonstruktionen, detta eftersom det är viktigt att bibehålla termisk noggrannhet från bilderna i modellen. Tre huvudresultat erhålls från denna avhandling. För det första är det möjligt att beräkna kamerakalibrering och position såväl som en gles rekonstruktion från en infraröd bildsekvens, detta med implementationen som föreslås i denna avhandling. För det andra presenteras och analyseras korrelationen för temperaturmätningar i bilderna som används för rekonstruktionen. Slutligen så visar den testade förbehandlingen inte en förbättring av rekonstruktionen som är propotionerlig med den ökade beräkningskomplexiteten.

When forensic examiners try to identify the perpetrator of a felony, they use individual facial marks when comparing the suspect with the perpetrator. Facial marks are often used for identification and they are nowadays found manually. To speed up this process, it is desired to detect interesting facial marks automatically. This master thesis describes a method to automatically detect and separate permanent and non-permanent marks. It uses a fast radial symmetry algorithm as a core element in the mark detector. After candidate skin mark extraction, the false detections are removed depending on their size, shape and number of hair pixels. The classification of the skin marks is done with a support vector machine and the different features are examined. The results show that the facial mark detector has a good recall while the precision is poor. The elimination methods of false detection were analysed as well as the different features for the classifier. One can conclude that the color of facial marks is more relevant than the structure when classifying them into permanent and non-permanent marks. / När forensiker försöker identifiera förövaren till ett brott använder de individuella ansiktsmärken när de jämför den misstänkta med förövaren. Dessa ansiktsmärken identifieras och lokaliseras oftast manuellt idag. För att effektivisera denna process, är det önskvärt att detektera ansiktsmärken automatiskt. I rapporten beskrivs en framtagen metod som möjliggör automatiskt detektion och separation av permanenta och icke-permanenta ansiktsmärken. Metoden som är framtagen använder en snabb radial symmetri algoritm som en huvuddel i detektorn. När kandidater av ansiktsmärken har tagits, elimineras alla falska detektioner utifrån deras storlek, form och hårinnehåll. Utifrån studiens resultat visar sig detektorn ha en god känslighet men dålig precision. Eliminationsmetoderna av falska detektioner analyserades och olika attribut användes till klassificeraren. I rapporten kan det fastställas att färgskiftningar på ansiktsmärkena har en större inverkan än formen när det gäller att sortera dem i permanenta och icke-permanenta märken.

Now and then train accidents occur. Collisions between trains and objects such as animals, humans, cars, and fallen trees can result in casualties, severe damage on the train, and delays in the train traffic. Thus, train collisions are a considerable problem with consequences affecting society substantially. The company Termisk Systemteknik AB has on commission by Rindi Solutions AB investigated the possibility to detect anomalies on the railway using a trainmounted thermal imaging camera. Rails are also detected in order to determine if an anomaly is on the rail or not. However, the rail detection method does not work satisfactory at long range. The purpose of this master’s thesis is to improve the previous rail detector at long range by using machine learning, and in particular deep learning and a convolutional neural network. Of interest is also to investigate if there are any advantages using cross-modal transfer learning. A labelled dataset for training and testing was produced manually. Also, a loss function tailored to the particular problem at hand was constructed. The loss function was used both for improving the system during training and evaluate the system’s performance during testing. Finally, eight different approaches were evaluated, each one resulting in a different rail detector. Several of the rail detectors, and in particular all the rail detectors using crossmodal transfer learning, perform better than the previous rail detector. Thus, the new rail detectors show great potential to the rail detection problem.

The recent emergence of time-of-flight cameras has opened up new possibilities in the world of computer vision. These compact sensors, capable of recording the depth of a scene in real-time, are very advantageous in many applications, such as scene or object reconstruction. This thesis first addresses the problem of fusing depth data with color images. A complete process to combine a time-of-flight camera with a color camera is described and its accuracy is evaluated. The results show that a satisfying precision is reached and that the step of calibration is very important. The second part of the work consists of applying super-resolution techniques to the time-of-flight camera in order to improve its low resolution. Different types of super-resolution algorithms exist but this thesis focuses on the combination of multiple shifted depth maps. The proposed framework is made of two steps: registration and reconstruction. Different methods for each step are tested and compared according to the improvements reached in term of level of details, sharpness and noise reduction. The results obtained show that Lucas-Kanade performs the best for the registration and that a non-uniform interpolation gives the best results in term of reconstruction. Finally, a few suggestions are made about future work and extensions for our solutions.

The objective of this master’s thesis work is to evaluate the potential benefit of a superpixel preprocessing step for general object detection in a traffic environment. The various effects of different superpixel parameters on object detection performance, as well as the benefit of including depth information when generating the superpixels are investigated. In this work, three superpixel algorithms are implemented and compared, including a proposal for an improved version of the popular Spectral Linear Iterative Clustering superpixel algorithm (SLIC). The proposed improved algorithm utilises a coarse-to-fine approach which outperforms the original SLIC for high-resolution images. An object detection algorithm is also implemented and evaluated. The algorithm makes use of depth information obtained by a stereo camera to extract superpixels corresponding to foreground objects in the image. Hierarchical clustering is then applied, with the segments formed by the clustered superpixels indicating potential objects in the input image. The object detection algorithm managed to detect on average 58% of the objects present in the chosen dataset. It performed especially well for detecting pedestrians or other objects close to the car. Altering the density distribution of the superpixels in the image yielded an increase in detection rate, and could be achieved both with or without utilising depth information. It was also shown that the use of superpixels greatly reduces the amount of computations needed for the algorithm, indicating that a real-time implementation is feasible.