Multi-target tracking is the process that allows the extraction of object motion patterns of interest from a scene. Motion patterns are often described through metadata representing object locations and shape information. In the first part of this thesis we discuss the state-of-the-art methods aimed at accomplishing this task on monocular views and also analyse the methods for evaluating their performance. The second part of the thesis describes our research contribution to these topics. We begin presenting a method for multi-target tracking based on track-before-detect (MTTBD) formulated as a particle filter. The novelty involves the inclusion of the target identity (ID) into the particle state, which enables the algorithm to deal with an unknown and unlimited number of targets. We propose a probabilistic model of particle birth and death based on Markov Random Fields. This model allows us to overcome the problem of the mixing of IDs of close targets. We then propose three evaluation measures that take into account target-size variations, combine accuracy and cardinality errors, quantify long-term tracking accuracy at different accuracy levels, and evaluate ID changes relative to the duration of the track in which they occur. This set of measures does not require pre-setting of parameters and allows one to holistically evaluate tracking performance in an application-independent manner. Lastly, we present a framework for multi-target localisation applied on scenes with a high density of compact objects. Candidate target locations are initially generated by extracting object features from intensity maps using an iterative method based on a gradient-climbing technique and an isocontour slicing approach. A graph-based data association method for multi-target tracking is then applied to link valid candidate target locations over time and to discard those which are spurious. This method can deal with point targets having indistinguishable appearance and unpredictable motion. MT-TBD is evaluated and compared with state-of-the-art methods on real-world surveillance.

Despite advances in image, video, and motion capture technologies, human interactions are frequently represented as line drawings. Intuitively, drawings provide a useful way of filtering complex, dynamic sequences to produce concise representations of interaction. They also make it possible to represent phenomena such as topic spaces, that do not have a concrete physical manifestation. However, the processes involved in producing these drawings, the advantages and limitations of line drawings as representations, and the implications of drawing as an analytic method have not previously been investigated. This thesis explores the use of drawings to represent human interaction and is informed by the prior experience and abilities of the investigator as a practising visual artist. It begins by discussing the drawing process and how it has been used to capture human activities. Key drawing techniques are identified and tested against an excerpt from an interaction between architects. A series of new drawings are constructed to depict one scene from this interaction, highlighting the contrasts between each drawing technique and their impact on the way shared spaces are represented. A second series of original drawings are produced exploring new ways of representing these spaces, leading to a proposal for a field-based approach that combines gesture paths, fields, and human figures to create a richer analytic representation. A protocol for using this approach to analyse video in practice is developed and evaluated though a sequence of three participatory workshops for researchers in human interaction. The results suggest that the field based process of drawing facilitates the production of spatially enriched graphical representations of qualitative spaces. The thesis concludes that the use of drawing to explore non-metric approaches to shared interactional space, has implications for research in human interaction, interaction design, clinical psychology, anthropology, and discourse analysis, and will find form in new new approaches to contemporary artistic practice.

Location estimation when deployed on wireless networks supports a range of services including user tracking and monitoring, health care support and push and pull marketing. The main subject of this thesis is improving indoor and outdoor location estimation accuracy using received signal strength (RSS) from neighbouring base stations (BSs) or access points (APs), without using the global positioning system (GPS) or triangulation methods. For the outdoor environment, state-of-the-art deterministic and probabilistic algorithms are adapted to exploit principal components (PCs) and clustering. The accuracy is compared with K-nearest neighbour (KNN) algorithms using different partitioning models. The proposed scheme clusters the RSS tuples based on deviations from an estimated RSS attenuation model and then transforms the raw RSS in each cluster into new uncorrelated dimensions, using PCs. As well as simple global dimensionality reduction using PCs, the data reduction and rotation within each cluster improves estimation accuracy because a) each cluster can model the different local RSS distributions and b) it efficiently preserves the RSS correlations that are observed (some of which are substantial) in local regions and which independence approximations ignore. Different simulated and real environments are used for the comparisons. Experimental results show that positioning accuracy is significantly improved and fewer training samples are needed compared with traditional methods. Furthermore, a technique to adjust RSS data so that radio maps collected in different environmental conditions can be used together to enhance accuracy is also demonstrated. Additionally, in the radio coverage domain, a non-parametric probability approach is used for the radio reliability estimation and a semi-supervised learning model is proposed for the monitoring model training and evolution according to real-time mobile users’ RSS feedback. For the indoor environment, an approach for a large multi-story indoor location estimaiii tion using clustering and rank order matching is described. The accuracies using WiFi RSS alone, cellular GSM RSS alone and integrated WiFi and GSM RSS are presented. The methods were tested on real indoor environments. A hierarchical clustering method is used to partition the RSS space, where a cluster is defined as a set of mobile users who share exactly the same strongest RSS ranking set of transmitters. The experimental results show that while integrating of WiFi RSS with GSM RSS creates a marginal improvement, the GSM data can be used to ameliorate the loss of accuracy when APs fail.

In the recent years, the rapid development of digital technologies and the low cost of recording media have led to a great increase in the availability of multimedia content worldwide. This availability places the demand for the development of advanced search engines. Traditionally, manual annotation of video was one of the usual practices to support retrieval. However, the vast amounts of multimedia content make such practices very expensive in terms of human effort. At the same time, the availability of low cost wearable sensors delivers a plethora of user-machine interaction data. Therefore, there is an important challenge of exploiting implicit user feedback (such as user navigation patterns and eye movements) during interactive multimedia retrieval sessions with a view to improving video search engines. In this thesis, we focus on automatically annotating video content by exploiting aggregated implicit feedback of past users expressed as click-through data and gaze movements. Towards this goal, we have conducted interactive video retrieval experiments, in order to collect click-through and eye movement data in not strictly controlled environments. First, we generate semantic relations between the multimedia items by proposing a graph representation of aggregated past interaction data and exploit them to generate recommendations, as well as to improve content-based search. Then, we investigate the role of user gaze movements in interactive video retrieval and propose a methodology for inferring user interest by employing support vector machines and gaze movement-based features. Finally, we propose an automatic video annotation framework, which combines query clustering into topics by constructing gaze movement-driven random forests and temporally enhanced dominant sets, as well as video shot classification for predicting the relevance of viewed items with respect to a topic. The results show that exploiting heterogeneous implicit feedback from past users is of added value for future users of interactive video retrieval systems.

OFDMA will be the multiple access scheme for next generation networks, including LTE and LTE-A. These networks will provide higher data rates than now, up to several hundred Mbps. These new networks, using a higher carrier frequency and offering flexible bandwidth for different application types, require advanced techniques for radio resource management. One approach that has been suggested to improve the radio resource management is to use smart and semi-smart antennas, so that the coverage of a certain cell can be divided into several adjustable sectors by using different antenna patterns. However, for a multi-cell environment, there is a need to prevent there being a gap between adjacent cells as the antenna patterns change. In this work, Genetic Algorithms are used to optimize the antenna patterns to get better coverage together with better cell throughput, at the same time making sure there are no gaps. This thesis not only considers the overall problem, but also investigates the suitability of the Genetic Algorithm itself, with it being optimized to improve the performance of the radio resource management in LTE networks. The influence of selection rate and mutation rate on GA is investigated and tested by simulation. These Genetic Algorithms are used in a model of multi-cell LTE networks to optimise subchannel allocation combined with dynamic sectorisation. Different types of scenarios are considered and the Genetic Algorithm is used to solve the problem of combining subchannel allocation with dynamic sectorisation to give the best overall performance of the LTE network.

No description available.

This thesis is about the design of software which enables groups of people to make music together. Networked musical interaction has been an important aspect of Sound and Music Computing research since the early days, although collaborative music software has yet to gain mainstream popularity, and there is currently limited research on the design of such interfaces. This thesis draws on research from Computer Supported Cooperative Work (CSCW) to explore the design of systems for Collaborative Digital Musical Interaction (CDMI). A central focus of this research is the concept of Awareness: a person’s understanding of what is happening, and of who is doing what. A novel software interface is developed and used over three experimental studies to investigate the effects different interface designs have on the way groups of musicians collaborate. Existing frameworks from CSCW are extended to accommodate the properties of music as an auditory medium, and theories of conventional musical interaction are used to elaborate on the nature of music making as a collaborative and social activity which is focused on process-oriented creativity. This research contributes to the fields of Human-Computer Interaction (HCI), Computer Supported Cooperative Work, and Sound and Music Computing through the identification of empirically derived design implications and recommendations for collaborative musical environments. These guidelines are demonstrated through the design of a hypothetical collaborative music system. This thesis also contributes towards the methodology for evaluating such systems, and considers the distinctions between CDMI and the forms of collaboration traditionally studied within CSCW.

No description available.

Music Information Retrieval (MIR) tasks, in the main, are concerned with the accurate generation of one of a number of different types of music metadata {beat onsets, or melody extraction, for example. Almost always, they operate on fully mixed digital audio recordings. Commonly, this means that a large amount of signal processing effort is directed towards the isolation, and then identification, of certain highly relevant aspects of the audio mix. In some cases, results of one MIR algorithm are useful, if not essential, to the operation of another { a chord detection algorithm for example, is highly dependent upon accurate pitch detection. Although not clearly defined in all cases, certain rules exist which we may take from music theory in order to assist the task { the particular note intervals which make up a specific chord, for example. On the question of generating accurate, low level music metadata (e.g. chromatic pitch and score onset time), a potentially huge advantage lies in the use of multitrack, rather than mixed, audio recordings, in which the separate instrument recordings may be analysed in isolation. Additionally, in MIR, as in many other research areas currently, there is an increasing push towards the use of the Semantic Web for publishing metadata using the Resource Description Framework (RDF). Semantic Web technologies, though, also facilitate the querying of data via the SPARQL query language, as well as logical inferencing via the careful creation and use of web ontology language (OWL) ontologies. This, in turn, opens up the intriguing possibility of deferring our decision regarding which particular type of MIR query to ask of our low-level music metadata until some point later down the line, long after all the heavy signal processing has been carried out. In this thesis, we describe an over-arching vision for an alternative MIR paradigm, built around the principles of early, studio-based metadata capture, and exploitation of open, machine-readable Semantic Web data. Using the specific example of structural segmentation, we demonstrate that by analysing multitrack rather than mixed audio, we are able to achieve a significant and quantifiable increase in the accuracy of our segmentation algorithm. We also provide details of a new multitrack audio dataset with structural segmentation annotations, created as part of this research, and available for public use. Furthermore, we show that it is possible to fully implement a pair of pattern discovery algorithms (the SIA and SIATEC algorithms { highly applicable, but not restricted to, symbolic music data analysis) using only SemanticWeb technologies { the SPARQL query language, acting on RDF data, in tandem with a small OWL ontology. We describe the challenges encountered by taking this approach, the particular solution we've arrived at, and we evaluate the implementation both in terms of its execution time, and also within the wider context of our vision for a new MIR paradigm.

Automatic Music Transcription seeks a machine understanding of a musical signal in terms of pitch-time activations. One popular approach to this problem is the use of spectrogram decompositions, whereby a signal matrix is decomposed over a dictionary of spectral templates, each representing a note. Typically the decomposition is performed using gradient descent based methods, performed using multiplicative updates based on Non-negative Matrix Factorisation (NMF). The final representation may be expected to be sparse, as the musical signal itself is considered to consist of few active notes. In this thesis some concepts that are familiar in the sparse representations literature are introduced to the AMT problem. Structured sparsity assumes that certain atoms tend to be active together. In the context of AMT this affords the use of subspace modelling of notes, and non-negative group sparse algorithms are proposed in order to exploit the greater modelling capability introduced. Stepwise methods are often used for decomposing sparse signals and their use for AMT has previously been limited. Some new approaches to AMT are proposed by incorporation of stepwise optimal approaches with promising results seen. Dictionary coherence is used to provide recovery conditions for sparse algorithms. While such guarantees are not possible in the context of AMT, it is found that coherence is a useful parameter to consider, affording improved performance in spectrogram decompositions.