Homography estimation is a fundamental task in many computer vision applications, but many techniques for estimation rely on complicated feature extraction pipelines. We extend research in direct homography estimation (i.e. without explicit feature extraction) by implementing a convolutional network capable of estimating homographies. Previous work in deep learning based homography estimation calculates homographies between pairs of images, whereas our network takes single image input and registers it to a reference view where no image data is available. The application of the work is registering frames from American football video to a top-down view of the field. Our model manages to register frames in a test set with an average corner error equivalent to less than 2 yards. / Homografiuppskattning är ett förkrav för många problem inom datorseende, men många tekniker för att uppskatta homografier bygger på komplicerade processer för att extrahera särdrag mellan bilderna. Vi bygger på tidigare forskning inom direkt homografiuppskattning (alltså, utan att explicit extrahera särdrag) genom att  implementera ett Convolutional Neural Network (CNN) kapabelt av att direkt uppskatta homografier. Arbetet tillämpas för att registrera bilder från video av amerikansk fotball till en referensvy av fotbollsplanen. Vår modell registrerar bildramer från ett testset till referensvyn med ett snittfel i bildens hörn ekvivalent med knappt 2 yards.

<div> Planar homography estimation is foundational to many computer vision problems, such as Simultaneous Localization and Mapping (SLAM) and Augmented Reality (AR). However, conditions of high variance confound even the state-of-the-art algorithms. In this report, we analyze the performance of two recently published methods using Convolutional Neural Networks (CNNs) that are meant to replace the more traditional feature-matching based approaches to the estimation of homography. Our evaluation of the CNN based methods focuses particularly on measuring the performance under conditions of significant noise, illumination shift, and occlusion. We also measure the benefits of training CNNs to varying degrees of noise. Additionally, we compare the effect of using color images instead of grayscale images for inputs to CNNs. Finally, we compare the results against baseline feature-matching based homography estimation methods using SIFT, SURF, and ORB. We find that CNNs can be trained to be more robust against noise, but at a small cost to accuracy in the noiseless case. Additionally, CNNs perform significantly better in conditions of extreme variance than their feature-matching based counterparts. With regard to color inputs, we conclude that with no change in the CNN architecture to take advantage of the additional information in the color planes, the difference in performance using color inputs or grayscale inputs is negligible. About the CNNs trained with noise-corrupted inputs, we show that training a CNN to a specific magnitude of noise leads to a ``Goldilocks Zone'' with regard to the noise levels where that CNN performs best.</div>

<p>The field of autonomous vehicle navigation and localization is a highly active research</p><p>topic. The aim of this thesis is to evaluate the feasibility to use outdoor visual navigation in a semi-structured environment. The goal is to develop a visual navigation system for an autonomous golf ball collection vehicle operating on driving ranges.</p><p>The image feature extractors SIFT and PCA-SIFT was evaluated on an image database</p><p>consisting of images acquired from 19 outdoor locations over a period of several weeks to</p><p>allow different environmental conditions. The results from these tests show that SIFT-type</p><p>feature extractors are able to find and match image features with high accuracy. The results also show that this can be improved further by a combination of a lower nearest neighbour threshold and an outlier rejection method to allow more matches and a higher ratio of correct matches. Outliers were found and rejected by fitting the data to a homography model with the RANSAC robust estimator algorithm. </p><p>A simulator was developed to evaluate the suggested system with respect to pixel noise from illumination changes, weather and feature position accuracy as well as the distance to features, path shapes and the visual servoing target image (milestone) interval. The system was evaluated on a total of 3 paths, 40 test combinations and 137km driven. The results show that with the relatively simple visual servoing navigation system it is possible to use mono-vision as a sole sensor and navigate semi-structured outdoor environments such as driving ranges.</p>

A laser triangulating camera system projects a laser line onto an object to create height curveson the object surface. By moving the object, height curves from different parts of the objectcan be observed and combined to produce a three dimensional representation of the object.The calibration of such a camera system involves transforming received data to get real worldmeasurements instead of pixel based measurements. The calibration method presented in this thesis focuses specifically on small fields ofview. The goal is to provide an easy to use and robust calibration method that can complementalready existing calibration methods. The tool should get as good measurementsin metric units as possible, while still keeping complexity and production costs of the calibrationobject low. The implementation uses only data from the laser plane itself making itusable also in environments where no external light exist. The proposed implementation utilises a complete scan of a three dimensional calibrationobject and returns a calibration for three dimensions. The results of the calibration havebeen evaluated against synthetic and real data.

Identification of individuals has been solved with many different solutions around the world, either using biometric data or external means of verification such as id cards or RFID tags. The advantage of using biometric measurements is that they are directly tied to the individual and are usually unalterable. Acquiring dependable measurements is however challenging when the individuals are uncooperative. A dependable system should be able to deal with this and produce reliable identifications. The system proposed in this thesis can autonomously classify uncooperative specimens from depth data. The data is acquired from a depth camera mounted in an uncontrolled environment, where it was allowed to continuously record for two weeks. This requires stable data extraction and normalization algorithms to produce good representations of the specimens. Robust descriptors can therefore be extracted from each sample of a specimen and together with different classification algorithms, the system can be trained or validated. Even with as many as 138 different classes the system achieves high recognition rates. Inspired by the research field of face recognition, the best classification algorithm, the method of fisherfaces, was able to accurately recognize 99.6% of the validation samples. Followed by two variations of the method of eigenfaces, achieving recognition rates of 98.8% and 97.9%. These results affirm that the capabilities of the system are adequate for a commercial implementation.

The field of autonomous vehicle navigation and localization is a highly active research topic. The aim of this thesis is to evaluate the feasibility to use outdoor visual navigation in a semi-structured environment. The goal is to develop a visual navigation system for an autonomous golf ball collection vehicle operating on driving ranges. The image feature extractors SIFT and PCA-SIFT was evaluated on an image database consisting of images acquired from 19 outdoor locations over a period of several weeks to allow different environmental conditions. The results from these tests show that SIFT-type feature extractors are able to find and match image features with high accuracy. The results also show that this can be improved further by a combination of a lower nearest neighbour threshold and an outlier rejection method to allow more matches and a higher ratio of correct matches. Outliers were found and rejected by fitting the data to a homography model with the RANSAC robust estimator algorithm. A simulator was developed to evaluate the suggested system with respect to pixel noise from illumination changes, weather and feature position accuracy as well as the distance to features, path shapes and the visual servoing target image (milestone) interval. The system was evaluated on a total of 3 paths, 40 test combinations and 137km driven. The results show that with the relatively simple visual servoing navigation system it is possible to use mono-vision as a sole sensor and navigate semi-structured outdoor environments such as driving ranges.

In this thesis, we present an algorithm for vision based detection of moving objects observed by IR sensors on a moving platform. In addition we analyze the performance of different approaches in each step of the algorithm. The proposed algorithm is composed of preprocessing, feature detection, feature matching, homography estimation and difference image analysis steps. First, a global motion estimation based on planar homography model is performed in order to compensate the motion of the sensor and moving platform where the sensors are located. Then, moving objects are identified on difference images of consecutive video frames with global motion suppression. Performance of the proposed algorithm is shown on different IR image sequences.

<p>A laser-triangulating range camera uses a laser plane to light an object. If the position of the laser relative to the camera as well as certrain properties of the camera is known, it is possible to calculate the coordinates for all points along the profile of the object. If either the object or the camera and laser has a known motion, it is possible to combine several measurements to get a three-dimensional view of the object.</p><p>Camera calibration is the process of finding the properties of the camera and enough information about the setup so that the desired coordinates can be calculated. Several methods for camera calibration exist, but this thesis proposes a new method that has the advantages that the objects needed are relatively inexpensive and that only objects in the laser plane need to be observed. Each part of the method is given a thorough description. Several mathematical derivations have also been added as appendices for completeness.</p><p>The proposed method is tested using both synthetic and real data. The results show that the method is suitable even when high accuracy is needed. A few suggestions are also made about how the method can be improved further.</p>

This degree project evaluates combinations of well-known state-of-the-art keypoint detectors and descriptors, as well as keypoint matching and robust outlier rejection methods for the purpose of estimating a homography between images produced by two fundamentally different cameras. The evaluation is perfomed on both computational efficiency and matching accuracy of each combination after a series of image deformations have been applied. The results show best performance using Brute Force search with the Hamming distance on keypoint descriptors generated by running the BRISK/BRISK combination and RANSAC for finding the subset to be used in the final homography estimation. If necessary for extra time sensitive applications, using ORB/ORB for keypoint detection and description has been shown to produce largely comparable results at a higher computational efficiency / Det här arbetet evaluerar kombinationer av välkända detektorer och deskriptorer, samt metoder för att matcha dessa och välja ut de bästa för att estimera en homografi mellan två fundamentalt olika kameror. Evalueringen baseras både på tidsåtgång och slutgiltig matchningskvalitet efter en rad bilddeformationer har applicerats. Resultaten visar bäst resultat när en totalsökning med Hammingnormen körs på ett set av punkter funna med hjälp av kombinationen BRISK/BRISK och sedan RANSAC för att hitta det bästa subsetet av dessa för att estimera homografin. Om nödvändigt för extra tidskänsliga applikationer har kombinationen ORB/ORB visat sig prestera i stort sett lika bra och med en ökad effektivitet.

This study investigated the significance of semantics in computer-generated word frequency counts in response to a call for new word lists (Read, 2000; Gardner, 2007). Read claims that no corpus projects to date have produced any "definitive, stand-alone word-frequency lists" (p. 226). Many researchers are wary of the fact that the concept of a word is never clearly defined in most studies that have dealt with word frequency counts. It is clear from the research that one universally acceptable construct for the concept of word does not exist. In fact, many past word frequency counts only examine word forms without considering the word meanings and the possible effects of homography on lists. Ming-Tzu and Nation (2004) did some research on the Academic Word List (AWL) that addresses some criticisms of word-frequency lists. They evaluate the extent of homography throughout the AWL. However, words found in the AWL are often not a part of the highest frequency word-forms in English. The present study focuses on high frequency words. It evaluates a randomized sample of 46 lemmas that occur at least 1500 times in the British National Corpus (BNC). A further random sampling of 200 examples for each lemma, in context, was semantically analyzed and tallied. One hundred of these examples were from the written portion and the other 100 from the spoken portion. The list of meanings for each word was compiled using conflated WordNet senses and some additional senses. Each context was double and sometimes triple rated. The results indicate that the impact of semantic frequency versus form-based frequency is considerable. The study suggests that the presence of homography tends to be extensive in many high-frequency word forms, across major registers of the language, and within each of the four major parts of speech. It further suggests that basing frequency on semantics will considerably alter the content of a high-frequency word list.