In this thesis, the Recursive-Random Sample Consensus (R-RANSAC) multiple target tracking (MTT) algorithm is further developed and applied to video taken from static platforms. Development of R-RANSAC is primarily focused in three areas: data association, the ability to track maneuvering objects, and track management. The probabilistic data association (PDA) filter performs very well in the R-RANSAC framework and adds minimal computation cost over less sophisticated methods. The interacting multiple models (IMM) filter as well as higher-order linear models are incorporated into R-RANSAC to improve tracking of highly maneuverable targets. An effective track labeling system, a more intuitive track merging criteria, and other improvements were made to the track management system of R-RANSAC. R-RANSAC is shown to be a modular algorithm capable of incorporating the best features of competing MTT algorithms. A comprehensive comparison with the Gaussian mixture probability hypothesis density (GM-PHD) filter was conducted using pseudo-aerial videos of vehicles and pedestrians. R-RANSAC maintains superior track continuity, especially in cases of interacting and occluded targets, and has fewer missed detections when compared with the GM-PHD filter. The two algorithms perform similarly in terms of the number of false positives and tracking precision. The concept of a feedback loop between the tracker and sensor processing modules is extensively explored; the output tracks from R-RANSAC are used to inform how video processing is performed. We are able to indefinitely detect stationary objects by zeroing out the background update rate of target-associated pixels in a Gaussian mixture models (GMM) foreground detector. False positive foreground detections are eliminated with a minimum blob area threshold, a ghost suppression algorithm, and judicious tuning of the R-RANSAC parameters. The ability to detect stationary targets also allows R-RANSAC to be applied to a class of problems known as stationary object detection. Additionally, moving camera foreground detection techniques are applied to the static camera case in order to produce measurements with a velocity component; this is accomplished by using sequential-RANSAC to cluster optical flow vectors of FAST feature pairs. This further improves R-RANSAC's track continuity, especially with interacting targets. Finally, a hybrid algorithm composed of R-RANSAC and the Sequence Model (SM), a machine learner, is presented. The SM learns sequences of target locations and is able to assist in data association once properly trained. In simulation, we demonstrate the SM's ability to significantly improve tracking performance in situations with infrequent measurement updates and a high proportion of clutter measurements.

Correct insertion of computer graphics into live-action broadcasts of outdoor sports requires precise knowledge of the foreground, i.e. players present in the scene. This thesis proposes a foreground detection and segmentation- framework with focus on real-time performance for 1080p resolution. A dataset consisting of four scenes; single-, multi-segment-, transcending-foreground and a light-witch scene all with dynamic backgrounds was constructed together with 26 ground-truths. Results show that the framework should run internally at 288p using GPU acceleration with geometrical nearest-neighbour-interpolation to attain real-time-capability. To maximize accuracy of the results, the framework uses two instances of OpenCV MOG2 in parallel on differently downsampled frames that are bitwise-joined to increase robustness. A set of morphological operations provides post-processing to get spatial coherence and a specific turf- consideration gives accurate contours. Thanks to additional camera- operator input, a crude distance-estimate lets foreground segments fade into background at a predetermined depth. The framework suffers from inaccurate segmentation during rapid light-switches, but recovers in a matter of seconds like the 'vanilla' MOG algorithm. For the specific scenes the framework provides excellent performance, especially considering the light-switch scene by comparison to the MOG-algorithm. For non-specific scenes of the 'BMC 2012' performance does not exceed the current state-of-the-art. / Korrekt placering av datorgrafik i video för tv-produktion kräver god känndedom om aktuell förgrund. Denna avhandling föreslår ett förgrundsdetektions- och segmenterings- ramverk med fokus på realtidsbearbetning av full-HD upplöst sport i utomhusmiljö. För utvärdering skapades ett dataset bestående av fyra scener; singel-, multisegment-, avlägsnande-förgrund och en ljusomväxlingsscen tillsammans med 26 referensförgrunder. För att erhålla realtidsbearbetning skall ramverket internt nyttja 288p upplösning med GPU acceleration och geometrisk närmaste-granne-interpolation. Resultaten visade att maximal noggranhet och ökad robusthet erhölls med två instanser av OpenCV MOG2 arbetandes parallellt på olikt nerskalade bilder för att därefter pixelvis förenas. För att erhålla sammanhängande förgrundssegment nyttjades morfologiska operationer på den binära sammansatta förgrunden vilket tillsammans med en specifik gräskantskorrektion ger precisa konturer. Tack vare givna kameraparametrar kan djupet till förgrundselementen uppskattas därmed låts de övergå till bakgrund för ett visst djupt. Ramverket lider av oprecis segmententering vid snabba ljusomväxlingar, men återhämtar sig när bakgrundsmodellen uppdaterats till de nya ljusförutsättningarna. För ovan nämnda specifika scener presterar ramverket utmärkt, speciellt med avseende på ljusomväxlingen, där prestandan är flerfaldigt bättre än den enskilda 'MOG'-metoden. För generella scener ur 'BMC 2012' datasetet presterar vår metod dock inte bättre än state-of-the-art.

Recent developments in the computer vision field achieve state-of-the-art performance by utilizing large-scale training datasets and in the absence of that, generating synthetic datasets of said magnitude. Yet, for certain applications, it is not feasible to synthesize high fidelity training data (e.g., biomedical computer vision domain), or to achieve detailed explainability for the program's decisions. Formulating a part-based approach can help alleviate the aforementioned challenges as (i) a scene can naturally be decomposed into a hierarchical part-based structure, and (ii) using domain knowledge by incorporating the object parts' topological and geometrical constraints reduces the complexity of learning and inference, benefiting methods in terms of data efficiency and computational resources. This dissertation investigates multiple applications that benefit from a part-based solution regarding the applications' performance metrics and/or computational efficiency. We develop part-based methods for registration, segmentation, unsupervised object discovery in large-scale image collections, and unsupervised unknown foreground discovery in streaming scenarios.

The thesis deals with an image processing problem called image matting. The problem involves detection of a foreground and background in an image with minimal user interaction using trimaps. Foreground detection is used in image composition. The goal of the thesis is to apply already known algorithms, in this case A Global sampling matting, in an Android application. The most important result is an intuitive application that can be used for making creative viral photos. Agile methodology is applied throughout the whole application development cycle. From the very beginning, the application is publicly available as a minimum viable product on Google play. The work’s contribution is in optimization of the mentioned algorithm for use in mobile devices and parallelization on a GPU, together with a publicly available application.

The effective surveillance of large critical areas is crucial for their security and privacy. There is no publicly available and acceptable solution of automating this task. This thesis aims to create an application utilizing a combination of a pan-tilt robotic manipulator and a visible-spectrum camera. Based on the pan-tilt unit's position and camera's images, the application searches for semantically significant changes in the captured environment and marks these regions of interest.