Video from aerial surveillance can provide a rich source of data for analysts. From the time-critical perspective of wilderness search and rescue operations, information extracted from aerial videos can mean the difference between a successful search and an unsuccessful search. When using low-cost, payload-limited mini-UAVs, as opposed to more expensive platforms, several challenges arise, including jittery video, narrow fields of view, low resolution, and limited time on screen for key features. These challenges make it difficult for analysts to extract key information in a timely manner. Traditional approaches may address some of these issues, but no existing system effectively addresses all of them in a unified and efficient manner. Building upon a hierarchical dense image correspondence technique, we create a unifying framework for reducing jitter, enhancing resolution, and expanding the field of view while lengthening the time that features remain on screen. It also provides for easy extraction of moving objects in the scene. Our method incorporates locally adaptive warps which allows for robust image alignment even in the presence of parallax and without the aid of internal or external camera parameters. We accelerate the image registration process using commodity Graphics Processing Units (GPUs) to accomplish all of these tasks in near real-time with no external telemetry data.

Cette thèse se concentre sur le développement demodèles de représentation dense d’objets 3-D àpartir d’images. L’objectif de ce travail estd’améliorer les modèles surfaciques 3-D fournispar les systèmes de vision par ordinateur, enutilisant de nouveaux éléments tirés des images,plutôt que les annotations habituellementutilisées, ou que les modèles basés sur unedivision de l’objet en différents parties.Des réseaux neuronaux convolutifs (CNNs) sontutilisés pour associer de manière dense les pixelsd’une image avec les coordonnées 3-D d’unmodèle de l’objet considéré. Cette méthodepermet de résoudre très simplement unemultitude de tâches de vision par ordinateur,telles que le transfert d’apparence, la localisationde repères ou la segmentation sémantique, enutilisant la correspondance entre une solution surle modèle surfacique 3-D et l’image 2-Dconsidérée. On démontre qu’une correspondancegéométrique entre un modèle 3-D et une imagepeut être établie pour le visage et le corpshumains. / This thesis addresses the task of establishing adense correspondence between an image and a 3Dobject template. We aim to bring vision systemscloser to a surface-based 3D understanding ofobjects by extracting information that iscomplementary to existing landmark- or partbasedrepresentations.We use convolutional neural networks (CNNs)to densely associate pixels with intrinsiccoordinates of 3D object templates. Through theestablished correspondences we effortlesslysolve a multitude of visual tasks, such asappearance transfer, landmark localization andsemantic segmentation by transferring solutionsfrom the template to an image. We show thatgeometric correspondence between an imageand a 3D model can be effectively inferred forboth the human face and the human body.

There is ongoing research into building dense correspondence between digital images of objects in the world and estimating the 3D pose of these objects. This is a difficult area to conduct research due to the lack of availability of annotated data. Annotating each pixel is too time-consuming. At the time of this writing, current research has managed to use neural networks to establish a dense pose estimation of human body parts (feet, chest, legs etc.). The aim of this thesis is to investigate if a model can be developed using neural networks to perform dense pose estimation on human feet. The data used in evaluating the model is generated using proprietary tools. Since this thesis is using a custom model and custom dataset, one model will be developed and tested with various experiments to gain an understanding of the different parameters that influence the model’s performance. Experiments showed that a model based on DeepLabV3 is able to achieve a dense pose estimation of feet with a mean error of 1.0cm. The limiting factor for a model’s ability to estimate a dense pose is based on the model’s ability to classify the pixels in an image accurately. It was also shown that discontinuous UV unwrapping greatly reduced the model’s dense pose estimation ability. The results from this thesis should be considered preliminary and need to be repeated multiple times to account for the stochastic nature of training neural networks. / Pågående forskning undersöker hur man kan skapa tät korrespondens mellan digitala bilder av objekt i världen och uppskatta de objektens 3D-pose. Detta är ett svårt område att forska inom på grund av bristen på tillgänglig annoterad data. Att annotera varje pixel är tidskrävande. Vid tiden för detta skrivande har aktuell forskning lyckats använda neurala nätverk för att etablera en tät pose-estimering av mänskliga kroppsdelar (fötter, bröst, ben osv.). Syftet med denna arbete är att undersöka om en modell kan utvecklas med hjälp av neurala nätverk för att utföra dense pose-estimering av mänskliga fötter. Data som används för att utvärdera modellen genereras med hjälp av proprietära verktyg. Eftersom denna arbete använder en anpassad modell och anpassad dataset kommer en modell att utvecklas och testas med olika experiment för att förstå de olika parametrarna som påverkar modellens prestanda. Experiment visade att en modell baserad på DeepLabV3 kan uppnå en dense pose-estimering av fötter med en medelfel på 1,0 cm. Den begränsande faktorn för en modells förmåga att uppskatta en dense pose baseras på modellens förmåga att klassificera pixlarna i en bild korrekt. Det visades också att oregelbunden UV-uppackning avsevärt minskade modellens förmåga att estimera dense pose. Resultaten från denna avhandling bör betraktas som preliminära och behöver upprepas flera gånger för att ta hänsyn till den stokastiska naturen hos träning av neurala nätverk.