DISCRETE OPTIMIZATION ALGORITHMS IN REAL-TIME VISUAL TRACKING

In this work we introduce a novel formulation of the association problem in visual tracking systems as a discrete optimization problem. The full data association problem is formulated as a search for the best tracking configuration to match hypothesis. We have implemented three local search algorithms: Hill Climbing, Simulated Annealing, and Tabu Search algorithms. These algorithms are guided by heuristic evaluation function which takes into account structural and specific information such as distance and shape. We have also introduced a novel technique in order to achieve incrementality in discrete optimization algorithms searching on an indirect space. We will show how this incrementality produces efficient and fast results, especially so when real-time is a hard constraint. The results obtained with the three discrete optimization algorithms are compared with other well-known and powerful computer vision tracking algorithms. We will prove the effectiveness and robustness of the discrete optimization algorithms in five different real-life scenarios.     

A method for online pattern recognition of abnormal eye movements

A doctor could say that a patient is sick while he/she is healthy or could say that the patient is healthy while he/she is sick, by mistake. So it is important to generate a system that can give a good diagnosis, in this case for abnormal eye movements. An abnormal eye movement is when the patient wants to move the eye to up or down and the eye does not move or the eye moves to other place. In this paper, a method for the pattern recognition is used to provide a better diagnosis for patients related with the abnormal eye movements. The real data of signals of two eye movements (up and down) of patients are obtained using a mindset ms-100 system. A new method that uses one intelligent algorithm for online pattern recognition is proposed. The difference between the proposed method and the previous works is that, in other works, both behaviors (up and down) are trained with one intelligent algorithm, while in this work, up behavior is trained with one intelligent algorithm and down behavior is trained with other intelligent algorithm; it is because the multi-output system can always be decomposed into a collection of single-output systems with the advantage to use different parameters for each one if necessary. The intelligent algorithm used by the proposed method could be any of the following: the adaline network denoted as AN, the multilayer neural network denoted as NN, or the Sugeno fuzzy inference system denoted as SF. So the comparison results of the proposed method using each of the intelligent algorithms for online pattern recognition of two eye movements are presented.     

Spider specie identification and verification based on pattern recognition of it cobweb

Biodiversity conservation is a global priority where the study of every type of living form is a fundamental task. Inside the huge number of the planet species, spiders play an important role in almost every habitat. This paper presents a comprehensive study on the reliability of the most used features extractors to face the problem of spider specie recognition by using their cobwebs, both in identification and verification modes. We have applied a preprocessing to the cobwebs images in order to obtain only the valid information and compute the optimal size to reach the highest performance. We have used the principal component analysis (PCA), independent component analysis (ICA), Discrete Cosine Transform (DCT), Wavelet Transform (DWT) and discriminative common vectors as features extractors, and proposed the fusion of several of them to improve the system’s performance. Finally, we have used the Least Square Vector Support Machine with radial basis function as a classifier. We have implemented K-Fold and Hold-Out cross-validation techniques in order to obtain reliable results. PCA provided the best performance, reaching a 99.65% ± 0.21 of success rate in identification mode and 99.98% ± 0.04 of the area under de Reveicer Operating Characteristic (ROC) curve in verification mode. The best combination of features extractors was PCA, DCT, DWT and ICA, which achieved a 99.96% ± 0.16 of success rate in identification mode and perfect verification.     

The application of nanoparticles in gene therapy and magnetic resonance imaging

The combination of nanoparticles, gene therapy, and medical imaging has given rise to a new field known as gene theranostics, in which a nanobioconjugate is used to diagnose and treat the disease. The process generally involves binding between a vector carrying the genetic information and a nanoparticle, which provides the signal for imaging. The synthesis of this probe generates a synergic effect, enhancing the efficiency of gene transduction and imaging contrast. We discuss the latest approaches in the synthesis of nanoparticles for magnetic resonance imaging, gene therapy strategies, and their conjugation and in vivo application.     

Fuzzy emerging patterns for classifying hard domains

Emerging pattern–based classification is an ongoing branch in Pattern Recognition. However, despite its simplicity and accurate results, this classification includes an a priori discretization step that may degrade the classification accuracy. In this paper, we introduce fuzzy emerging patterns as an extension of emerging patterns to deal with numerical attributes using fuzzy discretization. Based on fuzzy emerging patterns, we propose a new classifier that uses a novel graph organization of patterns. The new classifier outperforms some popular and state of the art classifiers on several UCI repository databases. In a pairwise comparison, it significantly beats every other single classifier.     

A Multi-agent Architecture Based on the BDI Model for Data Fusion in Visual Sensor Networks

The newest surveillance applications is attempting more complex tasks such as the analysis of the behavior of individuals and crowds. These complex tasks may use a distributed visual sensor network in order to gain coverage and exploit the inherent redundancy of the overlapped field of views. This article, presents a Multi-agent architecture based on the Belief-Desire-Intention (BDI) model for processing the information and fusing the data in a distributed visual sensor network. Instead of exchanging raw images between the agents involved in the visual network, local signal processing is performed and only the key observed features are shared. After a registration or calibration phase, the proposed architecture performs tracking, data fusion and coordination. Using the proposed Multi-agent architecture, we focus on the means of fusing the estimated positions on the ground plane from different agents which are applied to the same object. This fusion process is used for two different purposes: (1) to obtain a continuity in the tracking along the field of view of the cameras involved in the distributed network, (2) to improve the quality of the tracking by means of data fusion techniques, and by discarding non reliable sensors. Experimental results on two different scenarios show that the designed architecture can successfully track an object even when occlusions or sensor??s errors take place. The sensor??s errors are reduced by exploiting the inherent redundancy of a visual sensor network with overlapped field of views.     

Application of machine learning techniques to analyse student interactions and improve the collaboration process

In e-learning environments that use the collaboration strategy, providing participants with a set of communication services may not be enough to ensure collaborative learning. It is thus necessary to analyse collaboration regularly and frequently. Using machine learning techniques is recommended when analysing environments where there are a large number of participants or where they control the collaboration process. This research studied two approaches that use machine learning techniques to analyse student collaboration in a long-term collaborative learning experience during the academic years 2006–2007, 2007–2008 and 2008–2009. The aims were to analyse collaboration during the collaboration process and that it should be domain independent. Accordingly, the intention was to be able to carry out the analysis regularly and frequently in different collaborative environments. One of the two approaches classifies students according to their collaboration using unsupervised machine learning techniques, clustering, while the other approach constructs metrics that provide information on collaboration using supervised learning techniques, decision trees. The research results suggest that collaboration can be analysed in this way, thus achieving the aims set out with two different machine learning techniques.     

Beyond the structure of SAT formulas

Nowadays, many real-world problems are encoded into SAT instances and efficiently solved by modern SAT solvers. These solvers, usually known as Conflict-Driven Clause Learning (CDCL) SAT solvers, include a variety of sophisticated techniques, such as clause learning, lazy data structures, conflict-based adaptive branching heuristics, or random restarts, among others. However, the reasons of their efficiency in solving real-world, or industrial, SAT instances are still unknown. The common wisdom in the SAT community is that these technique exploit some hidden structure of real-world problems.In this thesis, we characterize some important features of the underlying structure of industrial SAT instances. Namely, they are the community structure and the self-similar structure. We observe that most industrial SAT formulas, viewed as graphs, have these two properties. This means that (i) in a graph with a clear community structure, i.e. having high modularity, we can find a partition of its nodes into communities such that most edges connect nodes of the same community; and (ii) in a graph with a self-similar pattern, i.e. being fractal, its shape is kept after re-scalings, i.e., grouping sets of nodes into a single node. We also analyze how these structures are affected by the effects of CDCL techniques during the search.Using the previous structural studies, we propose three applications. First, we face the problem of generating pseudo-industrial random SAT instances using the notion of modularity. Our model generates instances similar to (classical) random SAT formulas when the modularity is low, but when this value is high, our model is also adequate to model realistic pseudo-industrial problems. Second, we propose a method based on the community structure of the instance to detect relevant learnt clauses. Our technique augments the original instance with this set of relevant clauses, and this results into an overall improvement of the efficiency of several state-of-the-art CDCL SAT solvers. Finally, we analyze the classification of industrial SAT instances into families using the previously analyzed structure features, and we compare them to other classifiers commonly used in portfolio SAT approaches.In summary, this dissertation extends the understandings of the structure of SAT instances, with the aim of better explaining the success of CDCL techniques and possibly improve them, and propose a number of applications based on this analysis of the underlying structure of SAT formulas.