abstract: Detecting anatomical structures, such as the carina, the pulmonary trunk and the aortic arch, is an important step in designing a CAD system of detection Pulmonary Embolism. The presented CAD system gets rid of the high-level prior defined knowledge to become a system which can easily extend to detect other anatomic structures. The system is based on a machine learning algorithm --- AdaBoost and a general feature --- Haar. This study emphasizes on off-line and on-line AdaBoost learning. And in on-line AdaBoost, the thesis further deals with extremely imbalanced condition. The thesis first reviews several knowledge-based detection methods, which are relied on human being's understanding of the relationship between anatomic structures. Then the thesis introduces a classic off-line AdaBoost learning. The thesis applies different cascading scheme, namely multi-exit cascading scheme. The comparison between the two methods will be provided and discussed. Both of the off-line AdaBoost methods have problems in memory usage and time consuming. Off-line AdaBoost methods need to store all the training samples and the dataset need to be set before training. The dataset cannot be enlarged dynamically. Different training dataset requires retraining the whole process. The retraining is very time consuming and even not realistic. To deal with the shortcomings of off-line learning, the study exploited on-line AdaBoost learning approach. The thesis proposed a novel pool based on-line method with Kalman filters and histogram to better represent the distribution of the samples' weight. Analysis of the performance, the stability and the computational complexity will be provided in the thesis. Furthermore, the original on-line AdaBoost performs badly in imbalanced conditions, which occur frequently in medical image processing. In image dataset, positive samples are limited and negative samples are countless. A novel Self-Adaptive Asymmetric On-line Boosting method is presented. The method utilized a new asymmetric loss criterion with self-adaptability according to the ratio of exposed positive and negative samples and it has an advanced rule to update sample's importance weight taking account of both classification result and sample's label. Compared to traditional on-line AdaBoost Learning method, the new method can achieve far more accuracy in imbalanced conditions. / Dissertation/Thesis / M.S. Computing Studies 2011

abstract: Real-world environments are characterized by non-stationary and continuously evolving data. Learning a classification model on this data would require a framework that is able to adapt itself to newer circumstances. Under such circumstances, transfer learning has come to be a dependable methodology for improving classification performance with reduced training costs and without the need for explicit relearning from scratch. In this thesis, a novel instance transfer technique that adapts a "Cost-sensitive" variation of AdaBoost is presented. The method capitalizes on the theoretical and functional properties of AdaBoost to selectively reuse outdated training instances obtained from a "source" domain to effectively classify unseen instances occurring in a different, but related "target" domain. The algorithm is evaluated on real-world classification problems namely accelerometer based 3D gesture recognition, smart home activity recognition and text categorization. The performance on these datasets is analyzed and evaluated against popular boosting-based instance transfer techniques. In addition, supporting empirical studies, that investigate some of the less explored bottlenecks of boosting based instance transfer methods, are presented, to understand the suitability and effectiveness of this form of knowledge transfer. / Dissertation/Thesis / M.S. Computer Science 2011

The area of computer vision has been making big improvements in the latest decades, equally so has the area of electronics and small computers improved. These areas together have made it more available to build small, standalone systems for object detection in live video. This project's main objective is to examine whether a small device, e.g. Raspberry Pi 3, can manage an implementation of an object detection algorithm, called Viola-Jones, to count the occupancy of sitting people in a room with a camera. This study is done by creating an application with the library OpenCV, together with the language C+ +, and then test if the application can run on the small device. Whether or not the application will detect people depends on the models used, therefore three are tested: Haar Face, Haar Upper body and Haar Upper body MCS. The library's object detection function takes some parameters that works like settings for the detection algorithm. With that, the parameters needs to be tailored for each model and use case, for an optimal performance. A function was created to find the accuracy of different parameters by brute-force. The test showed that the Haar Face model was the most accurate. All the models, with their most optimal parameters, are then speed-tested with a FPS test on the raspberry pi. The result shows whether or not the raspberry pi can manage the application with the models. All models could be run and the Haar face model was fastest. As the system uses cameras, some ethical aspects are discussed about what people might think of top-corner cameras.

Submitted by Daniella Sodre (daniella.sodre@ufpe.br) on 2015-03-09T14:45:09Z No. of bitstreams: 2 dissertacao Andre Costa.pdf: 10722387 bytes, checksum: bff242b1a21e34e27f228538f8f5d6b1 (MD5) license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) / Made available in DSpace on 2015-03-09T14:45:10Z (GMT). No. of bitstreams: 2 dissertacao Andre Costa.pdf: 10722387 bytes, checksum: bff242b1a21e34e27f228538f8f5d6b1 (MD5) license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) Previous issue date: 2013-11-25 / Capes / Um detector de pedestres básico geralmente possui dois componentes principais: um que seleciona regiões da imagem que possivelmente contêm um pedestre (gerador de candidatos) e outro que classifica estas regiões em grupos de pedestres e não-pedestres (classificador). Estes classificadores normalmente baseiam-se em extratores de características, que são transformações que alteram a intensidade ou cor original dos pixels de uma imagem em uma nova representação, para ressaltar algum tipo de conhecimento sobre o conteúdo da imagem. Quando o ambiente é não-controlado, fatores externos podem influenciar negativamente no desempenho do classificador. Baixa resolução, ruído, desfoque e oclusão são alguns efeitos que podem ser gerados por estes fatores, degradando a qualidade das imagens obtidas e, consequentemente, das características extraídas. Esta dissertação propõe-se a avaliar como extratores de características comportam-se nesse tipo de ambiente. Estes cinco tipos de degradação foram simulados nas bases de imagem usadas nos experimentos: INRIA Person e Caltech Pedestrian. Como estamos interessados apenas na etapa de classificação, as imagens foram transformadas em janelas de tamanho fixo na etapa de pré-processamento. Os experimentos usam uma combinação de extratores de características (HOG, LBP, CSS, LGIP e LTP) e modelos de aprendizagem (AdaBoost e SVM linear) para formar classificadores. Os classificadores foram treinados com as imagens intactas e testados com imagens em diversos níveis de degradação. O HOG (42%) e LTP (54%) foram superiores aos demais em aproximadamente metade dos pontos de teste na INRIA Person e Caltech Pedestrian, respectivamente. Foi confirmada a queda de desempenho do LBP quando exposto a ruído, mostrando que o LGIP e o LTP amenizam isso. Também observou-se que o CSS é robusto a ruído, mas gera características fracas no geral. Por fim, notou-se que classificadores que combinam mais de um extrator de características foram superiores aos individuais em boa parte dos pontos de teste. Combinando-se todos os extratores, tem-se um classificador superior em 95,8% das situações ao criado somente com o melhor extrator no geral (HOG, na base da INRIA, e LTP, na base da Caltech).

The work presented in this thesis is part of a large research and development project on condition-based maintenance for heavy trucks and buses at Scania. The aim of this thesis was to be able to predict the status of a component (the starter motor) using data mining methods and to create models that can predict the failure of that component. Based on workshop history data, error codes and operational data, three sets of classification models were built and evaluated. The first model aims to find patterns in a set of error codes, to see which codes are related to a starter motor failure. The second model aims to see if there are patterns in operational data that lead to the occurrence of an error code. Finally, the two data sets were merged and a classifier was trained and evaluated on this larger data set. Two machine learning algorithms were used and compared throughout the model building: AdaBoost and random forest. There is no statistically significant difference in their performance, and both algorithms had an error rate around ~13%, ~5% and ~13% for the three classification models respectively. However, random forest is much faster, and is therefore the preferable option for an industrial implementation. Variable analysis was conducted for the error codes and operational data, resulting in rankings of informative variables. From the evaluation metric precision, it can be derived that if our random forest model predicts a starter motor failure, there is a 85.7% chance that it actually has failed. This model finds 32% (the models recall) of the failed starter motors. It is also shown that four error codes; 2481, 2639, 2657 and 2597 have the highest predictive power for starter motor failure classification. For the operational data, variables that concern the starter motor lifetime and battery health are generally ranked as important by the models. The random forest model finds 81.9% of the cases where the 2481 error code occurs. If the random forest model predicts that the error code 2481 will occur, there is a 88.2% chance that it will. The classification performance was not increased when the two data sets were merged, indicating that the patterns detected by the two first classification models do not add value toone another.

We propose a new ensemble algorithm: the meta-boosting algorithm. This algorithm enables the original Adaboost algorithm to improve the decisions made by different WeakLearners utilizing the meta-learning approach. Better accuracy results are achieved since this algorithm reduces both bias and variance. However, higher accuracy also brings higher computational complexity, especially on big data. We then propose the parallelized meta-boosting algorithm: Parallelized-Meta-Learning (PML) using the MapReduce programming paradigm on Hadoop. The experimental results on the Amazon EC2 cloud computing infrastructure show that PML reduces the computation complexity enormously while retaining lower error rates than the results on a single computer. As we know MapReduce has its inherent weakness that it cannot directly support iterations in an algorithm, our approach is a win-win method, since it not only overcomes this weakness, but also secures good accuracy performance. The comparison between this approach and a contemporary algorithm AdaBoost.PL is also performed.

This Master thesis deals with usage of radar signal for vehicle classification. The thesis uses radar modules with continuous wave based on Doppler effect. Radar signal is processed by a series of signal processing method finished by Fourier transform. Data produced by FFT is used to create SVM and AdaBoost classifier which can be used to classify vehicles into groups.

This master‘s thesis describes the identification faces in real terms. It includes an overview of current methods of detection faces by the classifiers. It also includes various methods for detecting faces. The second part is a description of two programs designed to identify persons. The first program operates in real time under laboratory conditions, where using web camera acquires images of user's face. This program is designed to speed recognition of persons. The second program has been working on static images, in real terms. The main essence of this method is successful recognition of persons, therefore the emphasis on computational complexity. The programs I used a staged method of PCA, LDA and kernel PCA (KPCA). The first program only works with the PCA method, which has good results with respect to the success and speed of recognition. In the second program to compare methods, which passed the best method for KPCA.

This diploma thesis is focused on detection in images using extended set of Haar-like features and histogram-based method. At first is introduced a basic concept of extraction and classification image features. The next part bring own concept of image features based on Diffusion distance. Result of this work is implementation this methods in Rapidminer.

This thesis deals with human body detection and gestures tracking in videosequences. First, processing of videosequences in general is described. Further, different methods of human body detection are described and represented by significant papers. The most of the attention is focused on detection by real AdaBoost algorithm based on Haar-like features and Edgelet features. The practical part starts with selection of method that is implemented in this thesis. This method is detection by real AdaBoost based on Haar-like features. Further, different options of videosequence processing in JAVA are researched with justification of choice OpenCV library with JavaCV wrapper, which is used in this thesis. In the end, application itself is described, including description of GUI and description of each class and its functionality.