This thesis explores problematics of speaker localization using microphone array. Aim of this thesis is implementation of algorithms for speaker localization and experiments with those algorithms. Calculation of TDOA was done using cross-correlation and hyperbolic method was used to calculate position estimation. Finished microphone array is able to locate speaker within certain variance. Results of this thesis allow reader to make assumptions regarding accuracy of localisation using microphone array and ARM kit with limited performance. Precision of position estimation using microphone array reached several decimeters, but this precision is dependent on distance from microphone array.

This thesis investigates the development of a low-cost passive acoustic system for localizing moving vessels to monitor areas where human activities such as fishing, snorkeling and poaching are restricted. The system uses several off-the-shelf sensors with unsynchronized clocks where the Time Difference of Arrival (TDOA) or time delay is extracted by cross-correlation of the signal between paired sensors. The cross-correlation function uses phase correlation or Phase Transform (PHAT) which whitens the cross-spectrum in order to de-emphasize dominant frequency components. Using the locations of pairs of sensors as foci, hyperbolic equations can be defined using the time delay between them. With three or more sensors, multiple hyperbolic functions can be calculated which intersect at a unique point: the boat's location. It is also found that increasing separation distances between sensors decreased the correlation between the signals. However larger separation distances have better localization capability than with small distances. Experimental results from the Columbia and Willamette Rivers are presented to demonstrate performance.

Thesis (Ph.D)--Electrical and Computer Engineering, Georgia Institute of Technology, 2009. / Committee Chair: Rusell M. Mersereau; Committee Member: Biing Hwang (Fred) Juang; Committee Member: Christopher E. Heil; Committee Member: Georgia Vachtsevanos; Committee Member: James H. McClellan. Part of the SMARTech Electronic Thesis and Dissertation Collection.

Within the last decade, a new type of signal acquisition has emerged called Compressive Sensing that has proven especially useful in providing a recoverable representation of sparse signals. This thesis presents similar results for Compressive Parametric Estimation. Here, signals known to lie on some unknown parameterized subspace may be recovered via randomized compressive measurements, provided the number of compressive measurements is a small factor above the product of the parametric dimension with the subspace dimension with an additional logarithmic term. In addition to potential applications that simplify the acquisition hardware, there is also the potential to reduce the computational burden in other applications, and we explore one such application in depth in this thesis. Source localization by matched-field processing (MFP) generally involves solving a number of computationally intensive partial differential equations. We introduce a technique that mitigates this computational workload by ``compressing'' these computations. Drawing on key concepts from the recently developed field of compressed sensing, we show how a low-dimensional proxy for the Green's function can be constructed by backpropagating a small set of random receiver vectors. Then, the source can be located by performing a number of ``short'' correlations between this proxy and the projection of the recorded acoustic data in the compressed space. Numerical experiments in a Pekeris ocean waveguide are presented which demonstrate that this compressed version of MFP is as effective as traditional MFP even when the compression is significant. The results are particularly promising in the broadband regime where using as few as two random backpropagations per frequency performs almost as well as the traditional broadband MFP, but with the added benefit of generic applicability. That is, the computationally intensive backpropagations may be computed offline independently from the received signals, and may be reused to locate any source within the search grid area. This thesis also introduces a round-robin approach for multi-source localization based on Matched-Field Processing. Each new source location is estimated from the ambiguity function after nulling from the data vector the current source location estimates using a robust projection matrix. This projection matrix effectively minimizes mean-square energy near current source location estimates subject to a rank constraint that prevents excessive interference with sources outside of these neighborhoods. Numerical simulations are presented for multiple sources transmitting through a generic Pekeris ocean waveguide that illustrate the performance of the proposed approach which compares favorably against other previously published approaches. Furthermore, the efficacy with which randomized back-propagations may also be incorporated for computational advantage (as in the case of compressive parametric estimation) is also presented.

This thesis represents Bayesian joint audio-visual tracking for the 3D locations of multiple people and a current speaker in a real conference environment. To achieve this objective, it focuses on several different research interests, such as acoustic-feature detection, visual-feature detection, a non-linear Bayesian framework, data association, and sensor fusion. As acoustic-feature detection, time-delay-of-arrival~(TDOA) estimation is used for multiple source detection. Localization performance using TDOAs is also analyzed according to different configurations of microphones. As a visual-feature detection, Viola-Jones face detection is used to initialize the locations of unknown multiple objects. Then, a corner feature, based on the results from the Viola-Jones face detection, is used for motion detection for robust objects. Simple point-to-line correspondences between multiple cameras using fundamental matrices are used to determine which features are more robust. As a method for data association and sensor fusion, Monte-Carlo JPDAF and a data association with IPPF~(DA-IPPF) are implemented in the framework of particle filtering. Three different tracking scenarios of acoustic source tracking, visual source tracking, and joint acoustic-visual source tracking are represented using the proposed algorithms. Finally the real-time implementation of this joint acoustic-visual tracking system using a PC, four cameras, and six microphones is addressed with two parts of system implementation and real-time processing.

Le nombre croissant de travaux réalisés ces dernières années a montré que la bioacoustique est particulièrement intéressante pour le suivi d’espèces discrètes. L’émergence de dispositifs d’enregistrement autonomes, associée à de nouvelles méthodes d’analyse, ont récemment participé à l’accroissement des études dans ce domaine. Au cours des 30 dernières années, le Loup gris (Canis lupus), mammifère carnivore aux mœurs discrètes connu pour ses hurlements de longue portée, a fait l’objet de nombreuses études acoustiques. Ces dernières visaient notamment à améliorer son suivi, qui s’avère complexe du fait des grandes capacités de déplacement des loups, de l’étendue de leurs territoires et de la diversité des milieux dans lesquels ils vivent. Cependant, la bioacoustique passive a jusqu’alors très peu été exploitée pour le suivi du Loup. C’est dans ce contexte que la présente thèse s’est organisée autour de trois axes de recherche. Les deux premiers axes portent sur l’apport de la bioacoustique passive pour le suivi du Loup gris en milieu naturel. En combinant des analyses acoustiques, statistiques et cartographiques, le premier objectif a été d’élaborer une méthode pour l’échantillonnage spatial de vastes zones d’étude, afin d’y détecter des hurlements de loups à l’aide de réseaux d’enregistreurs autonomes. Ce même dispositif a ensuite permis, dans un second temps, de tester la possibilité de localiser les loups grâce à leurs hurlements. Les expérimentations conduites en milieu de moyenne montagne (Massif des Vosges) et de plaine (Côtes de Meuse), sur deux zones d’étude de 30 km² et avec un réseau de 20 enregistreurs autonomes, ont permis de démontrer l’intérêt de la bioacoustique passive pour le suivi du Loup gris. En effet, près de 70% des émissions sonores (son synthétique aux propriétés similaires à celles de hurlements de loups) ont été détectés par au moins un enregistreur autonome en milieu de moyenne montagne et plus de 80% en milieu de plaine, pour des distances enregistreurs– source sonore atteignant respectivement plus de 2.7 km et plus de 3.5 km. Grâce à un modèle statistique et à un Système d’Information Géographique, la probabilité de détection des hurlements a pu être cartographiée sur les deux zones. En moyenne montagne, elle était forte à très forte (>0.5) sur 5.72 km² de la zone d’étude, contre 21.43 km² en milieu de plaine. Les sites d’émission ont été localisés avec une précision moyenne de 315 ± 617 (SD) m, réduite à 167 ± 308 (SD) m après l’application d’un seuil d’erreur temporelle défini d’après la distribution des données. Le troisième axe de travail porte quant à lui sur l’application d’indices de diversité acoustique pour estimer le nombre d’individus participant à un chorus et ainsi contribuer au suivi de l’effectif des meutes. Les valeurs obtenues pour les six indices (H, Ht, Hf, AR, M et ACI) étaient corrélées avec le nombre de loups hurlant dans les chorus artificiels testés. De bonnes prédictions de l’effectif ont été obtenues sur des chorus réels avec l’un de ces indices (ACI). L’influence de plusieurs biais sur la précision des prédictions de chacun des six indices a ensuite pu être étudiée, montrant que trois d’entre eux y étaient relativement peu sensibles (Hf, AR et ACI). Finalement, les résultats obtenus avec les enregistreurs autonomes montrent le potentiel des méthodes acoustiques passives pour la détection de la présence de loups mais aussi pour les localiser avec une bonne précision, dans des milieux contrastés et à de larges échelles spatiale et temporelle. L’utilisation des indices de diversité acoustique ouvre également de nouvelles perspectives pour l’estimation de l’effectif des meutes. Prometteuses, l’ensemble des méthodes émergeant de ce travail nécessite à présent quelques investigations complémentaires avant d’envisager une application concrète pour le suivi du Loup gris dans son milieu naturel / The growing number of studies carried out in recent years has shown that bioacoustics is particularly interesting for the monitoring of secretive species. The emergence of autonomous recording devices, combined with new methods of analysis, have recently contributed to the increase of studies in this field. Over the last 30 years, many bioacoustic studies have been developed for the Grey wolf (Canis lupus), a secretive large carnivore known for its howls spreading over distances up to several kilometers. These researches notably aimed to improve its monitoring, which is complex because of the strong wolf dispersal capacities over long distances, the large extent of their territories and the various natural contexts in which they live. In this context, this PhD thesis was organized around three research axes. The first two axes focused on the contribution of passive bioacoustics for the Grey wolf monitoring in the field. By combining acoustic, statistical and cartographic analysis, the first objective was to develop a spatial sampling method adapted to large study areas for the detection of wolf howls by using autonomous recorders. Then, the same protocol was used to investigate the possibility to localize wolves thanks to their howls. Field experimentations, conducted in mid-mountain (Massif des Vosges) and lowland (Côtes de Meuse) environments, in two study areas of 30 km² and with an array of 20 autonomous recorders, demonstrated the high potential of passive bioacoustics for the Grey wolf monitoring. Indeed, nearly 70% of broadcasts (synthetic sound with similar acoustic properties to howls) were detected by at least one autonomous recorder in mid-mountain environment and more than 80% in lowland environment, for sound source-recorders distances of up to 2.7 km and 3.5 km respectively. By using statistical model and Geographic Information System, the detection probability of wolf howls was modeled in both study areas. In the mid-mountain environment, this detection probability was high or very high (greater than 0.5) in 5.72 km² of the study area, compared with 21.43 km² in lowland environment. The broadcast sites were localized with an overall mean accuracy of 315 ± 617 (SD) m, reducing until 167 ± 308 (SD) m after setting a temporal error threshold defined from the data distribution. The third axe focused on the application of acoustic diversity indices to estimate the number of howling wolves in choruses and thus to contribute to pack size monitoring. Index values of the six indices (H, Ht, Hf, AR, M, and ACI) were positively correlated with the number of howling wolves in the artificial tested choruses. Interesting size predictions based on real choruses were obtained with one of the indices (ACI). The effects of several biases on the reference values for the acoustic indices were then explored, showing that three of them were relatively insensitive (Hf, AR and, ACI). Finally, results obtained with autonomous recorders confirm the real potential of passive acoustic methods for detecting the presence of wolves but also for localizing individuals with high precision, in contrasting natural environments, at large spatial and temporal scales. The use of acoustic diversity indices also opens new perspectives for estimating pack sizes. All of the promising methods emerging from this thesis require now further investigations before considering a concrete application for monitoring the Grey wolf in its natural environment