This thesis presents a method for completing target regions ("hole filling") in RGB stereo pairs. It builds upon the state of the art for completing single images by matching to and then blending source patches drawn from the rest of the image. A method is introduced for first completing the respective disparity maps using a coupled partial differential equation based on that of Bertalmio, et al. extended to create mutual disparity consistency. Estimated disparities are then used to guide completion of the missing color image texture. An extension to the coherence-based objective function introduced by Wexler, et al. is then introduced, which not only encourages coherence of the respective images with respect to source images but also stereoscopic consistency between the two. The PatchMatch algorithm of Barnes, et al. is extended to cross-image searching and matching. This matching is capable of automatically copying from corresponding unoccluded portions of the other image without requiring an explicit preliminary warping step. Stereoscopic consistency is produced by giving preference to matches with cross-image consistency when blending source patches. Additionally, the PatchMatch algorithm is extended to draw from scaled texture in a directed fashion based on the 3D structure of the scene estimated from the stereo image pairs. Results demonstrate that this method produces better completion than either single-image completion or previous methods for stereo completion.

Ändamålet med denna uppsats är att undersöka hur vi kan demonstrera en mer realistisk ljudbild genom högtalaren som återskapar ljudkällan. För att få en klarare bild av hur ett realistiskt återskapande av en ljudkälla skulle kunna fungera, genomfördes ett anonymt lyssningstest med sex stycken deltagare där de via en enkät fick delge sina tankar kring vad de upplevde som den mest realistiska återgivningen av en ljudkälla. Testet utformades på så sätt att testdeltagarna, med ögonbindel, först fick lyssna till en gitarrist som spelade en komposition, och därefter fick lyssna till en inspelad version av samma komposition genom tre stycken olika högtalarkonfigurationer: mono, stereo och enegenkonstruerad konfiguration med högtalare i fyra riktningar: vänster, höger, upp, ner. Alla testdeltagare har tidigare erfarenhet inom mixning och musikskapande för att lättare i ord kunna beskriva och delge vad det är för auditiva skillnader de upplever. För att kunna få utökad förståelse för hur vi kan återskapa ljud mer realistiskt har även en intervju genomförts med en ljudingenjör som har över 30 års erfarenhet i branschen med band som bland annat Red hot chili peppers. I denna studie användes en kvalitativ metod där varje testdeltagare har fått besvara frågorna i lugn och ro. Svaren från deltagarna samt intervjun, har analyserats för att kunna anknytas till tidigare forskning och närliggande teori. Detta för att kunna få en fördjupad förståelse kring hur vi upplever olika typer av ljudkällor samt vad realistisk återgivning av ljud är för lyssnaren. Undersökningens svar bidrar till ytterligare stöd för tidigare forskning om hur vi upplever ljudskillnader inom den psykoakustiska teorin inom ljud.

No description available.

With growing concerns about global food supply and environmental impacts of modern agriculture, we are seeing an increased demand for more horticultural research. While research into plant genetics has seen an increased throughput from recent technological advancements, plant phenotypic research throughput has lagged behind. Improvements in open-source image processing software and image capture hardware have created an opportunity for the development of more competitively-priced, faster data-acquisition tools. These tools could be used to collect measurements of plants' phenotype on a much larger scale without sacrificing data quality. This paper demonstrates the feasibility of creating such a tool. The resulting design utilized stereo vision and image processes in the OpenCV project to measure a representative collection of observable plant traits like leaflet length or plant height. After the stereo camera was assembled and calibrated, visual and stereo images of potato plant canopies and tubers(potatoes) were collected. By processing the visual data, the meaningful regions of the image (the canopy, the leaflets, and the tubers) were identified. The same regions in the stereo images were used to determine plant physical geometry, from which the desired plant measurements were extracted. Using this approach, the tool had an average accuracy of 0.15 inches with respect to distance measurements. Additionally, the tool detected vegetation, tubers, and leaves with average Dice indices of 0.98, 0.84, and 0.75 respectively. To compare the tool's utility to that of traditional implements, a study was conducted on a population of 27 potato plants belonging to 9 separate genotypes. Both newly developed and traditional measurement techniques were used to collect measurements of a variety of the plants' characteristics. A multiple linear regression of the plant characteristics on the plants' genetic data showed that the measurements collected by hand were generally better correlated with genetic characteristics than those collected using the developed tool; the average adjusted coefficient of determination for hand-measurements was 0.77, while that of the tool-measurements was 0.66. Though the aggregation of this platform's results is unsatisfactory, this work has demonstrated that such an alternative to traditional data-collection tools is certainly attainable. / Master of Science

Visual perception has been a significant subject matter of robotics research for decades but has accelerated in recent years as both technology and community are more prepared to take on new challenges with autonomous systems. In this thesis, a framework for 3D reconstruction using a stereo camera for the purpose of obstacle detection and mapping is presented. In this application, a UAV works collaboratively with a UGV to provide high level information of the environment by using a downward facing stereo camera. The approach uses frame to frame SURF feature matching to detect candidate points within the camera image. These feature points are projected into a sparse cloud of 3D points using stereophotogrammetry for ICP registration to estimate the rigid transformation between frames. The RTK-GPS constrained pose estimate from the UAV is fused with the feature matched estimate to align the reconstruction and eliminate drift. The reconstruction was tested on both simulated and real data. The results indicate that this approach improves frame to frame registration and produces a well aligned reconstruction for a single pass compared to using the raw UAV position estimate alone. However, multi-pass registration errors occur on the order of about 0.6 meters between parallel passes, and approximately 2 degrees of local rotation error when compared to a reconstruction produced with Agisoft Metashape. However, the proposed system performed at an average frame rate of about 1.3 Hz compared to Agisoft at 0.03 Hz. Overall, the system improved obstacle registration and can perform online within existing ROS frameworks. / Master of Science / Visual perception has been a significant subject matter of robotics research for decades but has accelerated in recent years as both technology and community are more prepared to take on new challenges with autonomous systems. In this thesis, a framework for 3D reconstruction using cameras for the purpose of obstacle detection and mapping is presented. In this application, a UAV works collaboratively with a UGV to provide high level information of the environment by using a downward facing stereo camera. The approach uses features extracted from camera images to detect candidate points to be aligned. These feature points are projected into a sparse cloud of 3D points using stereo triangulation techniques. The 3D points are aligned using an iterative solver to estimate the translation and rotation between frames. The RTK (Real Time Kinematic) GPS constrained position and orientation estimate from the UAV is combined with the feature matched estimate to align the reconstruction and eliminate accumulated errors. The reconstruction was tested on both simulated and real data. The results indicate that this approach improves frame to frame registration and produces a well aligned reconstruction for a single pass compared to using the raw UAV position estimate alone. However, multi-pass registration errors occur on the order of about 0.6 meters between parallel passes that overlap, and approximately 2 degrees of local rotation error when compared to a reconstruction produced with the commercial product, Agisoft. However, the proposed system performed at an average frame rate of about 1.3 Hz compared to Agisoft at 0.03 Hz. Overall, the system improved obstacle registration and can perform online within existing Robot Operating System frameworks.

In the emerging arena of face-to-face collaboration using large, wall-size screens, a good videoconferencing system would be useful for two locations which both have a large screen. But as screens get bigger, a single camera becomes less than adequate to drive a videoconferencing system for the entire screen. Even if a wide-angle camera is placed in the center of the screen, it's possible for people standing at the sides to be hidden. We can fix this problem by placing several cameras evenly distributed in a grid pattern (what we call a sparse camera array) and merging the photos into one image. With a single camera, people standing near the sides of the screen are viewing an image with a viewpoint at the middle of the screen. Any perspective projection used in this system will look distorted when standing at a different viewpoint. If an orthogonal projection is used, there will be no perspective distortion, and the image will look correct no matter where the viewer stands. As a first step in creating this videoconferencing system, we use stereo matching to find the real world coordinates of objects in the scene, from which an orthogonal projection can be generated. / Master of Science

In order to interact with the real world, humans need to perform several tasks such as object detection, pose estimation, motion estimation and distance estimation. These tasks are all part of scene understanding and are fundamental tasks of computer vision. Depth estimation received unprecedented attention from the research community in recent years due to the growing interest in its practical applications (ie robotics, autonomous driving, etc.) and the performance improvements achieved with deep learning. In fact, the applications expanded from the more traditional tasks such as robotics to new fields such as autonomous driving, augmented reality devices and smartphones applications. This is due to several factors. First, with the increased availability of training data, bigger and bigger datasets were collected. Second, deep learning frameworks running on graphical cards exponentially increased the data processing capabilities allowing for higher precision deep convolutional networks, ConvNets, to be trained. Third, researchers applied unsupervised optimization objectives to ConvNets overcoming the hurdle of collecting expensive ground truth and fully exploiting the abundance of images available in datasets. This thesis addresses several proposals and their benefits for unsupervised depth estimation, i.e., (i) learning from resynthesized data, (ii) adversarial learning, (iii) coupling generator and discriminator losses for collaborative training, and (iv) self-improvement ability of the learned model. For the first two points, we developed a binocular stereo unsupervised depth estimation model that uses reconstructed data as an additional self-constraint during training. In addition to that, adversarial learning improves the quality of the reconstructions, further increasing the performance of the model. The third point is inspired by scene understanding as a structured task. A generator and a discriminator joining their efforts in a structured way improve the quality of the estimations. Our intuition may sound counterintuitive when cast in the general framework of adversarial learning. However, in our experiments we demonstrate the effectiveness of the proposed approach. Finally, self-improvement is inspired by estimation refinement, a widespread practice in dense reconstruction tasks like depth estimation. We devise a monocular unsupervised depth estimation approach, which measures the reconstruction errors in an unsupervised way, to produce a refinement of the depth predictions. Furthermore, we apply knowledge distillation to improve the student ConvNet with the knowledge of the teacher ConvNet that has access to the errors.

This paper describes two methods for constructing a depth map from images. Each method has two stages. First, one or more needle maps are determined using a pair of images. This process employs either the Marr-Poggio-Grimson stereo and shape-from-shading, or, instead, photometric stereo. Secondly, a depth map is constructed from the needle map or needle maps computed by the first stage. Both methods make use of an iterative relaxation method to obtain the final depth map.

Recovering three-dimensional information from two-dimensional images is the fundamental goal of stereo techniques. The problem of recovering depth (three-dimensional information) from a set of images is essentially the correspondence problem: Given a point in one image, find the corresponding point in each of the other images. Finding potential correspondences usually involves matching some image property. If the images are from nearby positions, they will vary only slightly, simplifying the matching process. Once a correspondence is known, solving for the depth is simply a matter of geometry. Real images are composed of noisy, discrete samples, therefore the calculated depth will contain error. This error is a function of the baseline or distance between the images. Longer baselines result in more precise depths. This leads to a conflict: short baselines simplify the matching process, but produce imprecise results; long baselines produce precise results, but complicate the matching process. In this paper, we present a method for generating dense depth maps from large sets (1000's) of images taken from arbitrary positions. Long baseline images improve the accuracy. Short baseline images and the large number of images greatly simplifies the correspondence problem, removing nearly all ambiguity. The algorithm presented is completely local and for each pixel generates an evidence versus depth and surface normal distribution. In many cases, the distribution contains a clear and distinct global maximum. The location of this peak determines the depth and its shape can be used to estimate the error. The distribution can also be used to perform a maximum likelihood fit of models directly to the images. We anticipate that the ability to perform maximum likelihood estimation from purely local calculations will prove extremely useful in constructing three dimensional models from large sets of images.

Detta examensarbete är genomfört i samarbete med Thorbjörn Birging. Birging arbetar med RFID[1] och är även VD för CombiQ på Science Park i Jönköping. Under lång tid har han velat skapa en produkt som hjälper människor med sömnproblem, depression, ångest och stress. Författarens roll som ingenjörsstudent blev att utveckla hans ide till en färdig produkt. Birging upptäckte efter att ha läst en artikel om sömnproblem att det finns studier som visar att det är enklare att somna med ett svagt ljud i rummet för att andra personer i rummet inte skall störas. Birging fokuserade på att skapa en ny ljudkudde med denna teori som grund.  Målet med detta examensarbete blev att skapa en kudde med fokus på utmärkt stereoljud som hjälpredskap för människor med sömnproblem. Med ljudets användning blir insomningen betydligt lättare. Många människor söker hjälp för sina problem på sjukhus, men ofta får de bara recept på lugnande och antidepressiva mediciner. Ett annat problem med sömntabletter är risken för att utveckla ett beroende och en längre tids användning av medicinerna kan störa kroppens normala sömnfunktion, vilket läkaren Anders Isaksson som driver en privatklinik i Örebro kan intyga. Birging har haft kontakt med Anders under detta arbete och kan intyga detta.  Ett sätt att lindra sömnproblem är genom att patienten får lyssna till lugnande musik före insomnandet, vilket alltså är tanken med denna kudde. Författaren genomförde en undersökning för att ta reda på vad som är bra och dåligt med dagens ljudkuddar. En idégenerering och kravspecifikation påbörjades med hjälp av Birging. För att kunna veta vad kunderna önskar sig med denna nya produkt gjordes bland annat en designbrief, funktionsanalys och en QFD-matris. Designbriefen skapades både verbalt och visuellt för att se hur den nya produkten skall se ut på markanden. QFD-matrisen rangordnar kraven och konkurrenternas kuddar i förhållande till Birgings krav på den nya prototypen. QFD ger inte svar på vad kunden får för upplevelse med produkten vilket designbriefen kan åstadkomma. Konstruerandet av prototypen kan börja först när designbriefen och funktionsanalysen är klara. Materialvalet gjordes genom testfaser som ljudåtergivning och komfort. Skumplast materialen kommer ifrån Recticel AB i Gislaved och ett intresse för denna nya produkt finns i den framtida utvecklingen.  Denna nya produkt blev betydligt bättre än de ljudkuddar som finns på marknaden. Den nya produkten har egenskaper som tryckavlastning samt bättre ljudkvalitet vilket innebär att volymen inte behöver vara hög, vilket gör att omgivningen inte störs. Orsaken till att ljudet inte stör är att spridningen av ljudet mestadels riktas upp mot användaren. Kuddens mått är B50*H40*T5 cm innebär att man enkelt kan köpa vanliga örngott och att den inte ta onödig plats i sängen. [1] Radio Frequency IDentification / This exam is done with the cooperation of Thorbjörn Birging. Birging does work with RFID[1] and he is also CEO at CombiQ at Science Park in Jönköping. For a long time he wanted to make an innovation to help people with sleeping problems, depression, anxiety and stress. The author’s role as an engineer student where to make his idea come true from scratch to prototype. Birging discovered after reading an article about sleeping problems that there were studies which showed that it is easier to sleep with a faint sound in the room so other people in the room does not get disturbed. Birging then focused to create a new pillow with sound integrated that should stand out from today’s pillows.  The goal with this exam was to create a pillow with excellent stereo sound with amplifiers, which help peoples sleeping problems. Many people with these problems seek help at hospitals, but they only get prescriptions for tranquillizer and anti-depressive pills. Another issue with sleeping pills is the risk of getting addicted and long time use might disturb the normal sleeping function, which Anders Isaksson can confirm. Birging He runs a private clinic in Örebro and has helped Birging in a previous project. Birging has been in contact with Anders under this project and can confirm this statement.  One way of relive sleeping problems might be to let the patient listen to calming music before the sleep period begins, which is the idea for this project. The author did an investigation to find any positives or negatives in today’s pillows with sound ability. Idea generating and require specification started soon after this with help of Birging. To know what the customers want with this new product a designbrief, function analytic and a QFD-matrix. The designbrief where made both verbal and visual to know how the product should look like as a customer. With the QFD-matrix we can rank the preferences with the competitors pillows in relation to Birgings requirements on the new prototype. The QFD doesn’t give any answers how the customer experience the product that the designbrief can accomplish. A prototype construction can only begin when the designbrief and function analytic are finished. The material selection was made by different test-phases with sound and comfort. Recticel AB in Gislaved has provided the author with different types of foam for the pillow and for this new product there is an interest to follow the development in the future.   The new developed product did manage to become better than the competitor’s pillows in many ways. The new product is pressure relieving, gives better sound which means the volume does not have to be high, and disturbance in the surroundings is minimized. The reason is that the material does spread the sound waves straight upwards the users ears. The pillows measurement are B50*H40*T5, that means you can have much space in bed and easy find a standard pillowcase. The user will get a comfortably sleeping period the entire night with this new product. [1] Radio Frequency IDentification