No description available.

No description available.

No description available.

No description available.

Dissertação (mestrado)—Universidade de Brasília, Faculdade de Tecnologia, Departamento de Engenharia Elétrica, 2016. / Conforme o sangue flui através do corpo de um indivíduo, ele muda a forma como a luz é irradiada pela pele, pois o sangue absorve luz de forma diferente dos outros tecidos. Essa sutil variação pode ser capturada por uma câmera e ser usada para monitorar a atividade cardíaca de uma pessoa. O sinal capturado pela câmera é uma onda que representa as variações de tonalidade da pele ao longo do tempo. A frequência dessa onda é a mesma frequência na qual o coração bate. Portanto, o sinal capturado pela câmera pode ser usado para estimar a taxa cardíaca de uma pessoa. Medir o pulso cardíaco remotamente traz mais conforto pois evita o uso de eletrodos. Também permite o monitoramento de uma pessoa de forma oculta para ser empregado em um detector de mentira, por exemplo. Neste trabalho nós propomos dois algoritmos para a estimação da taxa cardíaca sem contato usando câmeras convencionais sob iluminação não controlada. O primeiro algoritmo proposto é um método simples que emprega um detector de face que identifica a face da pessoa sendo monitorada e extrai o sinal gerado pelas mudanças no tom da pele devido ao fluxo sanguíneo. Este algoritmo emprega um filtro adaptativo para aumentar a energia do sinal de interesse em relação ao ruído. Nós mostramos que este algoritmo funciona muito bem para vídeos com pouco movimento. O segundo algoritmo que propomos é uma melhora do primeiro para torná-lo mais robusto a movimentos. Nós modificamos o método usado para definir a região de interesse. Neste algoritmo é utilizado um detector de pele para eliminar pixels do plano de fundo do vídeo, os frames dos vídeos são divididos em micro-regiões que são rastreados com um algoritmo de fluxo ótico para compensar os movimentos e um algoritmo de clusterização é aplicado para selecionar automaticamente as melhores micro-regiões para efetuar a estimação da taxa cardíaca. Propomos também um esquema de filtragem temporal e espacial para reduzir o ruído introduzido pelo algoritmo de fluxo ótico. Comparamos os resultados dos nossos algoritmos com um oxímetro de dedo comercial e mostramos que eles funcionam bem para situações desafiadoras. / As the blood flows through the body of an individual, it changes the way that light is irradiated by the skin, because blood absorbs light differently than the remaining tissues. This subtle variation can be captured by a camera and be used to monitor the heart activity of a person. The signal captured by the camera is a wave that represents the changes in skin tone along time. The frequency of this wave is the same as the frequency by which the heart beats. Therefore, the signal captured by the camera could be used to estimate a person’s heart rate. This remote measurement of cardiac pulse provides more comfort as it avoids the use of electrodes or others devices attached to the body. It also allows the monitoring of a person in a canceled way to be employed in lie detectors, for example. In this work we propose two algorithms for non-contact heart rate estimation using conventional cameras under uncontrolled illumination. The first proposed algorithm is a simple approach that uses a face detector to identify the face of the person being monitored and extract the signal generated by the changes in the skin tone due to the blood flow. This algorithm employs an adaptive filter to boost the energy of the interest signal against noise. We show that this algorithm works very well for videos with little movement. The second algorithm we propose is an improvement of the first one to make it more robust to movements. We modify the approach used to define the region of interest. In this algorithm we employ a skin detector to eliminate pixels from the background, divide the frames in microregions that are tracked using an optical flow algorithm to compensate for movements and we apply a clustering algorithm to automatically select the best micro-regions to use for heart rate estimation. We also propose a temporal and spatial filtering scheme to reduce noise introduced by the optical flow algorithm. We compared the results of our algorithms to an off-the-shelf fingertip pulse oximeter and showed that they can work well under challenging situations.

This thesis presents methods and results for optimizing subword detectors in continuous speech. Speech detectors are useful within areas like detection-based ASR, pronunciation training, phonetic analysis, word spotting, etc. Firstly, we propose a structure suitable for subword detection. This structure is based on the standard HMM framework, but in each detector the MFCC feature extractor and the models are trained for the specific detection problem. Our experiments in the TIMIT database validate the effectiveness of this structure for detection of phones and articulatory features. Secondly, two discriminative training techniques are proposed for detector training. The first one is a modification of Minimum Classification Error training. The second one, Minimum Detection Error training, is the adaptation of Minimum Phone Error to the detection problem. Both methods are used to train HMMs and filterbanks in the detectors, isolated or jointly. MDE has the advantage that any detection performance criterion can be optimized directly. F-score and class accuracy optimization experiments show that MDE training is superior to the MCE-based method. The optimized filterbanks reflect some acoustical properties of the detection classes. Moreover, some changes are consistent over classes with similar acoustical properties. In addition, MDE-training of filterbanks results in filters significatively different than in the standard filterbank. In fact, some filters extract information from different critical bands. Finally, we propose a detection-based automatic speech recognition system. Detectors are built with the proposed HMM-based detection structure and trained discriminatively. The linguistic merger is based on an MLP/Viterbi decoder.

The purpose of this research is to design and optimize the performance of a unique, inexpensive 6Li neutron-capture pulse-mode ion chamber (LiPMIC) for neutron detection that overcomes the fill-gas contamination stemming from outgas of detector components. This research also provides a demonstration of performance of LiPMICs. Simulations performed with GARFIELD, a drift-chamber simulation package for ion transport in an electrostatic field, have shown that argon-methane mixtures of fill-gas allow maintenance of electron drift velocity through a surprisingly wide range of fill-gas content. During the design stage of LiPMIC development, the thicknesses of lithium metallization layer, the neutron energy conversion site of the detector, and the thickness of neutron moderator, the high-density polyethylene body, are optimized through analytical and MCNPX calculations. Also, a methodology of obtaining the suitable combination of electric field strength, electron drift velocity, and fill-gas mixtures has been tested and simulated using argon-methane gas mixtures. The LiPMIC is shown to have comparable efficiency to 3He proportional counters at a fraction of cost. Six-month long baseline measurements of overall detector performance shows there is a 3% reduction in total counts for 252Cf sources, which provides a good indicator for the longevity of the detector.

Events of significant impact as recent as hurricane Ike yielded a consistent disturbing truth: we lack sufficient numbers of competent search dog [Canis familiaris] teams. This study was conceived to provide information in identifying factors involved in training competent search dogs. Obedience training methods, age training was initiated, previous handler canine training experience, and handler perception and emotional attachment to their search dog were examined through a sixty-six question survey. Achievement of a national certification was used as a measure of performance success. Association between factors and performance success was evaluated through Chi-Square testing. Surveys were announced through the National Search Dog Alliance (NSDA) and were available online; 177 were fully completed by respondents and used in the data analysis. Seventy-two percent of nationally certified canine team respondents preferred positive reinforcement methods. Several statistically significant associations were detected: (a) female handlers preferred positive reinforcement training methods [x^2 = 8.504, d.f.=1, P = 0.004], (b) as dogs matured use of active training equipment increased [x^2 = 54.043, d.f.= 2, P < 0.001]., and (c) four hours or more time spent training each week had a higher proportion of national certifications [x^2 = 16.379, d.f. = 1, P less than 0.001]. The data also indicated a trend for handlers to have previous canine training experience equal to or greater than search dog training experience [x^2 = 118.36, d.f. = 9, P =0.05]. The results warrant further research on the effects of early training, the effects of training time investment, and the interaction between canine selection and handler understanding of canine learning theory.

N,N-Dimethylformamide (DMF) is a well-known chemical entity that is extensively used for pharmaceutical, biomedical and chemical applications. Previous research identified the need for the development of an effective dosage form for the systemic delivery of DMF due to its unique antiviral properties. For purposes of quality control and evaluation during pharmaceutical product development, development of an analytical method was required. A gas chromatographic (GC) method was developed with a flame-ionization detector (FID) on a carbowax packed glass column. 2-Methoxyethanol was used as internal standard. The analytical method proved to be capable of separating DMF and 2-methoxyethanol adequately within a relatively short runtime of 2.5 min. The analytical method described was primarily developed for use in dissolution studies of DMF containing delivery systems. Various physicochemical properties of candidate internal standard materials were correlated with the observed retention times of these compounds. The best correlation (r2 ¼ 0.8077) was obtained between the boiling point and the retention time of the compounds for the current application. The boiling point of an internal standard candidate material may therefore be useful in predicting the retention time of that compound under similar conditions.

DEAP (Dark Matter Experiment with Argon and Pulse Shape Discrimination) is an experiment that aims to directly detect dark matter particles via nuclear recoils in liquid argon. The experiment uses the scintillation property of liquid argon as a means to discriminate the γ and β backgrounds from the expected signal. DEAP-1 is a 7 kg single phase liquid argon detector. It was constructed to demonstrate the scalability for a larger (3600 kg) detector. The detector was originally operated at Queen’s University, where the background rejection level achieved was 6.3×10−8 for the recoil detection efficiency of 97.1%. The detector was relocated to SNOLAB, where the background in the energy region of interest was reduced by a factor of 7.7 (from 4.61±0.17 mHz to 0.60±0.05 mHz.). The background rejection level of 9.64×10−9 (10.4 part per billion) was achieved from the combined data set (Queen’s University and SNOLAB) for a recoil detection efficiency of 35.5 ± 1.3 %. With the current background rate, the background rejection level required for the 3600 kg detector (1.8×10−9 ) is projected to be achieved in 382 days at the neutron efficiency of 9.1±0.6 %. / Thesis (Master, Physics, Engineering Physics and Astronomy) -- Queen's University, 2009-01-26 09:28:25.432