Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / Esta dissertação apresenta um aperfeiçoamento para o Sistema de Imagens Tridimensional Híbrido (SITH) que é utilizado para obtenção de uma superfície tridimensional do relevo de uma determinada região a partir de dois aerofotogramas consecutivos da mesma. A fotogrametria é a ciência e tecnologia utilizada para obter informações confiáveis a partir de imagens adquiridas por sensores. O aperfeiçoamento do SITH consistirá na automatização da obtenção dos pontos através da técnica de Transformada de Características Invariantes a Escala (SIFT - Scale Invariant Feature Transform) dos pares de imagens estereoscópicas obtidos por câmeras aéreas métricas, e na utilização de técnicas de interpolação por splines cúbicos para suavização das superfícies tridimensionais obtidas pelo mesmo, proporcionando uma visualização mais clara dos detalhes da área estudada e auxiliando em prevenções contra deslizamentos em locais de risco a partir de um planejamento urbano adequado. Os resultados computacionais mostram que a incorporação destes métodos ao programa SITH apresentaram bons resultados. / This dissertation presents an improvement of the Three-Dimensional Hybrid Imaging System (SITH) that is used to obtain a three dimensional surface relief of a particular region from the same two consecutive air frames it. Photogrammetry is the science and technology used to obtain reliable information from images acquired by sensors. Improving the automation of the SITH will consist of points obtained using the technique of Invariant Feature Transform Scale (SIFT - Scale Invariant Feature Transform) pair of stereoscopic images obtained by aerial metric cameras, and the use of means of cubic spline interpolation for smooth surfaces produced by the same three-dimensional, providing a clearer view of the details of the study area and aiding in effective prevention against landslides in hazardous locations from an urban planning appropriate. The computational results show that the incorporation of these methods to the program SITH had good results.

Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / Esta dissertação apresenta um aperfeiçoamento para o Sistema de Imagens Tridimensional Híbrido (SITH) que é utilizado para obtenção de uma superfície tridimensional do relevo de uma determinada região a partir de dois aerofotogramas consecutivos da mesma. A fotogrametria é a ciência e tecnologia utilizada para obter informações confiáveis a partir de imagens adquiridas por sensores. O aperfeiçoamento do SITH consistirá na automatização da obtenção dos pontos através da técnica de Transformada de Características Invariantes a Escala (SIFT - Scale Invariant Feature Transform) dos pares de imagens estereoscópicas obtidos por câmeras aéreas métricas, e na utilização de técnicas de interpolação por splines cúbicos para suavização das superfícies tridimensionais obtidas pelo mesmo, proporcionando uma visualização mais clara dos detalhes da área estudada e auxiliando em prevenções contra deslizamentos em locais de risco a partir de um planejamento urbano adequado. Os resultados computacionais mostram que a incorporação destes métodos ao programa SITH apresentaram bons resultados. / This dissertation presents an improvement of the Three-Dimensional Hybrid Imaging System (SITH) that is used to obtain a three dimensional surface relief of a particular region from the same two consecutive air frames it. Photogrammetry is the science and technology used to obtain reliable information from images acquired by sensors. Improving the automation of the SITH will consist of points obtained using the technique of Invariant Feature Transform Scale (SIFT - Scale Invariant Feature Transform) pair of stereoscopic images obtained by aerial metric cameras, and the use of means of cubic spline interpolation for smooth surfaces produced by the same three-dimensional, providing a clearer view of the details of the study area and aiding in effective prevention against landslides in hazardous locations from an urban planning appropriate. The computational results show that the incorporation of these methods to the program SITH had good results.

No description available.

Image Registration plays a very important role in the field of remote sensing. In order to have a better registration quality and make the automatization possible, choos ing and matching the control points from conjugate images become very important. In fact, the control points required for image registration should have following three key factors, that is, the amount, validity and distribution of control points. ¡@¡@In the study, we take QuickBird Satellite Images as the main ones; on the other hand, it conducts two groups of image registrations resulted from aerial images at various times. After detecting feature points using different algorithms, the study makes use of feature matching methods to get conjugate points between two overlapped images. The algorithms used above are SIFT, ASIFT and MESR. SIFT is an algorithm which invariant to scales, rotation, affine stretch and change in brightness. ASIFT undertakes simulations based on the theory of SIFT and thus carries out fully affine invariant. The feature points obtained from MSER have physical meaning in its location. By using feature matching algorithms like K-d tree and BBF, the matched feature points from two overlapped images would be turned into the conjugate points which can be control points for image registration. ¡@¡@During the process of image preprocessing, it is learned that the feature points detected by SIFT and MSER through feature matching are very few. Hence, this study attempts to employ histogram specification¡Bcontrast stretching and scale change methods to see if it is helpful to the feature detections and matching through change of image quality and image size. The experiment found that scale change will improve both the amount and accuracy of conjugate points detected by different algorithms. When considering distribution of the feature points, the study takes advantage of image cropping approach to conduct feature detections and matching individually. It is found that more conjugate points with uniform distribution can be obtained via image cropping technique.

In this thesis, an image-based uncalibrated visual servo is proposed for image tracking tasks in highly disturbed environment, such as a remotely operated vehicle performing observing or investigation objects under the influence of undersea current. For the conditions that the target model and the camera parameters are unknown, the control framework applies the scale invariant feature transform (SIFT) to extract image features. Furthermore, a robust adaptive control law is implemented to overcome the effect caused by camera calibration parameters. Then by using three different types of camera¡¦s motion: pan, tilt, and zoom to maintain the target always at the central position on the image plane.

Tedlar™ bags are a promising medium for remote breath collection and later analysis using SIFT-MS for disease diagnosis. It is important to understand the changes in integrity of samples stored in Tedlar™ bags. However, there is little work into this problem completed to date, and thus little known about these issues. Therefore, a study into the integrity of samples stored in Tedlar™ bags and analysed using SIFT-MS was undertaken. The sample integrity of ammonia, acetone, ethanol, isoprene and pentane, all initially at 3ppm in breath and nitrogen substrates, and stored in Tedlar™ bags was investigated. Experiments tested the effect of storage size (0.5, 1, 3L), storage time (6-48 hours), storage temperature (23℃ - 25℃, 37℃), humidity (0.4 - 4.5% absolute) and inter-bag variation using triplicate bags. The SIFT-MS instrument used was LDI2 located at Christchurch Hospital. The repeatability and precision of LDI2 was established using prepared cylinder samples (0.05% absolute humidity) of acetone, pentane and ethanol tested at seven times over a 250 min time period. A generalised Cauchy distribution was used to give a combined distribution from multiple bags for the sample humidity and compound concentration. A combined measure of the repeatability and precision, T s , ranged between 217 - 349 ppb for ethanol, acetone and pentane. The factors affecting the repeatability and precision were both machine and compound dependant. The effect of the factors differed over time, with different precursors and compounds. No obvious effects of bag storage size on the sample integrity of pentane, isoprene, ethanol and acetone were observed. The absolute humidity change within bag samples was linked to the volume to surface area ratio because it was more affected by permeation and condensation. All compounds in the nitrogen substrate (except for 37℃ stored acetone (NO+)) displayed decreases in sample integrity with time. All compounds in the breath substrate displayed regular losses of sample integrity, except for the 37℃ and 23℃ - 25℃ stored ethanol (NO+) and 37℃ stored ethanol (H3O+), pentane (O2+) and ammonia (H3O+, O2+). The average change of sample integrity for pentane, isoprene, ethanol and acetone ranged from 0.2 to 3.6 times the maximum T s , while ammonia ranged from 0.9 - 10 times. All observed behaviour was reproducible. Absolute humidity and storage temperature affected the sample integrity of acetone, ethanol and ammonia. Generally, the intra-bag variance was comparable between all storage temperatures and substrates while the inter-bag variation was affected by the absolute humidity. Only the initial and final concentrations between precursors for the 23℃ - 25℃ stored breath and nitrogen substrates agreed. The breath substrate samples gave erroneous values for ammonia. Permeation of compounds into the bags from the atmosphere was not significant. The overall issues surrounding storing breath in Tedlar™ bags for analysis using SIFTMS is not the loss of sample integrity, but the kinetics, precision and repeatability of the SIFT-MS instrument. The current kinetics are not adequate to accurately monitor acetone, isoprene, pentane, ammonia and ethanol in breath and stored in Tedlar™ bags at breath absolute humidity levels greater than 3%. Generally, the loss of sample integrity was only marginally outside the repeatability and precision of the machine.

The usage of Unmanned Aerial Vehicles (UAV) for several applications has in-creased during the past years. One of the possible applications are aerial imagecapturing for detection and surveillance purposes. In order to make the captur-ing process more efficient, multiple, cameraequipped UAV:s could fly in a for-mation and as a result cover a larger area. To be able to receive several imagesequences and stitch those together, resulting in a panoramavideo, a softwareapplication has been developed and tested for this purpose.All functionality are developed in the language C++ by using the software li-brary OpenCV. All implementations of different techniques and methods hasbeen done as generic as possible to be able to add functionality in the future.Common methods in computervision and object recognition such as SIFT, SURF and RANSAC have been tested. / Användningen av UAV:er för olika tillämpningar har ökat under senare år. Ett avmöjliga användningsområden är flygfotografering och övervakning. För att gö-ra bildupptagningen mer effektiv kan flera UAV:er flyga i formation och på så viskunna fotografera ett avsevärt större område. För att kunna ta in flera bildsekven-ser och foga samman dessa till en panoramavideo, har ett program utvecklats ochtestats för denna uppgift.All funktionalitet för inläsning av bilder och video har utvecklats i C++ medprogrambiblioteket OpenCV. Implementeringen av dessa tekniker och metoderhar gjort så generiskt som möjligt för att det ska vara lättare att lägga till and-ra tekniker och utöka programmets funktioner. Olika tekniker som har testatsinkluderar: SIFT, SURF och RANSAC

This thesis presents several descriptor selection schemes for video content pairwise-matching tasks. Those proposed schemes attempt to leverage two significant properties of videos, temporal correlation and motion information. Aiming to find an efficient and descriptive representation for a video sequence, the concept of descriptor persistency is defined. Those descriptors that satisfy this definition are called persistent descriptors. In order to exploit descriptor persistency, an encoder is proposed. The proposed encoder consists of five main components. First, keyframe labelling is introduced to reduce complexity and ensure a reasonable size of persistent sets. After that, persistent descriptor detection is performed on each group of pictures (GOP) separately. The second component is the standard SIFT descriptor extraction. The third part is to identify persistent descriptors from each GOP, called persistent descriptors extraction. In this stage, three different methods are proposed: The direct method and two approximation approaches. Persistent descriptors selection, which is the fourth stage, is carried out to control the size of the persistent set. For this stage, three selection methods are proposed. All of them attempt to utilize the motion information to select more descriptive descriptors among all the persistent descriptors in the GOP. In order to perform pairwise-matching, in this thesis, a simple but efficient pairwise-matching method is proposed. Experiments are carried out to evaluate the performance of the proposed schemes. The datasets used for performance evaluation are subsets from the categories that describe in [1]. Two metrics de ned in [2], namely false positive rate (FPR) and true positive rate (TPR), are used for the performance evaluation. / Thesis / Master of Applied Science (MASc)

As técnicas de super-resolução possibilitam combinar várias imagens de uma mesma cena para se obter uma imagem com resolução radiométrica e geométrica aumentada, denominada de imagem de super-resolução. Nessa imagem são realçadas características importantes possibilitando recuperar detalhes e informações. As aplicações envolvem diferentes áreas, tais como: na agricultura para identificar possíveis desmatamentos e controle de pragas, na área médica para a detecção de doenças em estágios iniciais, identificação facial de pessoas suspeitas em imagens de circuito fechado, reconstrução de filmes, identificação de placas de veículos, entre outras. No presente trabalho, é proposta uma metodologia para a geração de imagens de super-resolução a partir de uma região selecionada de um mosaico. Embora existam vários trabalhos publicados relacionados à geração de imagens de super-resolução, as metodologias não se aplicam para uma região específica do mosaico. E grande parte dos trabalhos utiliza uma imagem de referência, a partir da qual é gerada a imagem de super-resolução. Na metodologia proposta, inicialmente, é gerado um mosaico a partir de um conjunto de imagens baseando-se nos algoritmos SIFT ou SURF, BBF e RANSAC e é criada uma estrutura de dados, que organiza os pontos correlacionáveis das imagens com sobreposição, facilitando e simplificando o processo de fusão desses pontos para a obtenção da imagem de super-resolução. A ferramenta implementada a partir dessa metodologia, possibilita ao operador selecionar a região de interesse no mosaico, a partir da qual, é gerada a imagem de super-resolução utilizando as técnicas SIFT (ou SURF), interpolação Bicúbica e a fusão pelo valor mediano dos pontos da área com sobreposição das imagens da sequência. Para validar a metodologia, foram utilizados quatro conjuntos de imagens, que incluem imagens simuladas, obtidas com câmeras de baixa e alta resolução, imagens aéreas de áreas urbanas e rurais, coloridas e em escalas de cinza, e imagens contendo elementos textuais. Nas imagens simuladas foram adicionados ruídos e avaliada a imagem de super-resolução gerada por meio de duas métricas: raiz do erro médio quadrático (RMSE) e o índice de similaridade estrutural (SSIM). Os resultados mostraram que mesmo com valor de RMSE elevado, o SSIM foi acima de 70%, evidenciando o alto grau de similaridade. As imagens de super-resolução obtidas a partir de uma região dos mosaicos gerados foram comparadas com imagens superamostradas por meio de interpolações e avaliadas confrontando as imagens extrapoladas para verificação visual dos elementos da cena. Os resultados apresentados concluem que as imagens de super-resolução geradas, apresentam melhorias no que diz respeito à restauração das mesmas para futura análise de alvos de interesse, sem ter o retrabalho de adquirir novas imagens da cena, pois dependendo da cena analisada não será possível nova aquisição. O presente trabalho contribui com a geração de imagem de super-resolução, a partir de uma região do mosaico e com estruturas de dados e algoritmos que possibilitam a análise de regiões específicas do mosaico, sem que o mesmo tenha que ser processado integralmente. / Super-resolution techniques allow combining several images of the same scene in order to obtain an image with increased geometric and radiometric resolution, called super-resolution image. In this image are enhanced features allowing to recover important details and information. The applications involve different areas, such as: in the agriculture to identify possible deforestation and pest control, in the medical area to detect diseases in early stages, facial identification in images of suspects in closed loop, movies reconstruction, license plates recognition, among others. In this work, we propose a methodology for generating super-resolution images from a selected region of a mosaic. Although there are several published papers related to the generation of super-resolution images, the existing methodologies do not apply to a specific region of the mosaic. And the majority of studies use a reference image, from which is generated the super-resolution image. In the proposed methodology, initially, a mosaic is generated from a set of images based on the algorithms SIFT or SURF, BBF and RANSAC and creates a data structure that organizes the points correlate of the images with overlapping, facilitating and simplifying the fusion process of these points to obtain the super-resolution image. The tool implemented from this methodology allows to the operator to select the region of interest in the mosaic, from which is generated the image using super-resolution techniques SIFT (or SURF), Bicubic interpolation and fusion process by the median value of the points with overlapping area from the images of the sequence. In order to validate the methodology, we used four sets of images, including simulated images taken with cameras of low and high resolution, aerial images of urban and rural areas, color and grayscale images and images containing texts. In the simulated images were added noise and were evaluated the super-resolution image generated by two metrics: root mean square error (RMSE) and the structural similarity index (SSIM). The results showed that even with high RMSE value, the SSIM was above 70%, reflecting the high degree of similarity. The super-resolution images obtained from a region of the mosaics were compared with images generated by super-sampled interpolation and evaluated by comparing the images extrapolated to visual inspection of elements of the scene. From the results it can be concluded that the super-resolution images generated present improvements with regard to restoration of images for further analysis of targets of interest, without reworking to acquire new images of the scene, because depending on the analyzed scene it would not be possible a new acquisition. This work contributes to the generation of super-resolution image from a region of the mosaic and with data structures and algorithms which enable the analysis of specific regions of the mosaic without it has to be fully processed.

As técnicas de super-resolução possibilitam combinar várias imagens de uma mesma cena para se obter uma imagem com resolução radiométrica e geométrica aumentada, denominada de imagem de super-resolução. Nessa imagem são realçadas características importantes possibilitando recuperar detalhes e informações. As aplicações envolvem diferentes áreas, tais como: na agricultura para identificar possíveis desmatamentos e controle de pragas, na área médica para a detecção de doenças em estágios iniciais, identificação facial de pessoas suspeitas em imagens de circuito fechado, reconstrução de filmes, identificação de placas de veículos, entre outras. No presente trabalho, é proposta uma metodologia para a geração de imagens de super-resolução a partir de uma região selecionada de um mosaico. Embora existam vários trabalhos publicados relacionados à geração de imagens de super-resolução, as metodologias não se aplicam para uma região específica do mosaico. E grande parte dos trabalhos utiliza uma imagem de referência, a partir da qual é gerada a imagem de super-resolução. Na metodologia proposta, inicialmente, é gerado um mosaico a partir de um conjunto de imagens baseando-se nos algoritmos SIFT ou SURF, BBF e RANSAC e é criada uma estrutura de dados, que organiza os pontos correlacionáveis das imagens com sobreposição, facilitando e simplificando o processo de fusão desses pontos para a obtenção da imagem de super-resolução. A ferramenta implementada a partir dessa metodologia, possibilita ao operador selecionar a região de interesse no mosaico, a partir da qual, é gerada a imagem de super-resolução utilizando as técnicas SIFT (ou SURF), interpolação Bicúbica e a fusão pelo valor mediano dos pontos da área com sobreposição das imagens da sequência. Para validar a metodologia, foram utilizados quatro conjuntos de imagens, que incluem imagens simuladas, obtidas com câmeras de baixa e alta resolução, imagens aéreas de áreas urbanas e rurais, coloridas e em escalas de cinza, e imagens contendo elementos textuais. Nas imagens simuladas foram adicionados ruídos e avaliada a imagem de super-resolução gerada por meio de duas métricas: raiz do erro médio quadrático (RMSE) e o índice de similaridade estrutural (SSIM). Os resultados mostraram que mesmo com valor de RMSE elevado, o SSIM foi acima de 70%, evidenciando o alto grau de similaridade. As imagens de super-resolução obtidas a partir de uma região dos mosaicos gerados foram comparadas com imagens superamostradas por meio de interpolações e avaliadas confrontando as imagens extrapoladas para verificação visual dos elementos da cena. Os resultados apresentados concluem que as imagens de super-resolução geradas, apresentam melhorias no que diz respeito à restauração das mesmas para futura análise de alvos de interesse, sem ter o retrabalho de adquirir novas imagens da cena, pois dependendo da cena analisada não será possível nova aquisição. O presente trabalho contribui com a geração de imagem de super-resolução, a partir de uma região do mosaico e com estruturas de dados e algoritmos que possibilitam a análise de regiões específicas do mosaico, sem que o mesmo tenha que ser processado integralmente. / Super-resolution techniques allow combining several images of the same scene in order to obtain an image with increased geometric and radiometric resolution, called super-resolution image. In this image are enhanced features allowing to recover important details and information. The applications involve different areas, such as: in the agriculture to identify possible deforestation and pest control, in the medical area to detect diseases in early stages, facial identification in images of suspects in closed loop, movies reconstruction, license plates recognition, among others. In this work, we propose a methodology for generating super-resolution images from a selected region of a mosaic. Although there are several published papers related to the generation of super-resolution images, the existing methodologies do not apply to a specific region of the mosaic. And the majority of studies use a reference image, from which is generated the super-resolution image. In the proposed methodology, initially, a mosaic is generated from a set of images based on the algorithms SIFT or SURF, BBF and RANSAC and creates a data structure that organizes the points correlate of the images with overlapping, facilitating and simplifying the fusion process of these points to obtain the super-resolution image. The tool implemented from this methodology allows to the operator to select the region of interest in the mosaic, from which is generated the image using super-resolution techniques SIFT (or SURF), Bicubic interpolation and fusion process by the median value of the points with overlapping area from the images of the sequence. In order to validate the methodology, we used four sets of images, including simulated images taken with cameras of low and high resolution, aerial images of urban and rural areas, color and grayscale images and images containing texts. In the simulated images were added noise and were evaluated the super-resolution image generated by two metrics: root mean square error (RMSE) and the structural similarity index (SSIM). The results showed that even with high RMSE value, the SSIM was above 70%, reflecting the high degree of similarity. The super-resolution images obtained from a region of the mosaics were compared with images generated by super-sampled interpolation and evaluated by comparing the images extrapolated to visual inspection of elements of the scene. From the results it can be concluded that the super-resolution images generated present improvements with regard to restoration of images for further analysis of targets of interest, without reworking to acquire new images of the scene, because depending on the analyzed scene it would not be possible a new acquisition. This work contributes to the generation of super-resolution image from a region of the mosaic and with data structures and algorithms which enable the analysis of specific regions of the mosaic without it has to be fully processed.