Traditional object detection methods struggle in complex logistics and warehousing environments due to their inability to effectively identify regions of interest while handling extreme occlusion, high object density, and significant scale variations with static. Existing region proposal approaches (anchor-based, anchor-free, and deep learning-based) require meticulous hyperparameter tuning, struggle with small or overlapping objects, have difficulties with different lighting conditions or suffer from poor localization in cluttered scenes. We introduce Adaptive Tiles , a novel preprocessing algorithm that dynamically identifies potential object regions prior to detection, thereby addressing these limitations. The core mechanism involves a three-step process: (1) zero-shot segmentation for adaptable, object-agnostic mask generation, (2) filter heuristics to refine masks into robust region proposals, and (3) bounding box combination for merging proposals into optimal detection zones. By shifting from static region identification methods to a dynamic, segmentation-driven proposal mechanism, Adaptive Tiles significantly enhances the efficiency and accuracy of object detectors, as demonstrated by its integration into the DMZoomNet framework and evaluation on the challenging LOCO dataset.

This paper introduces a new system for mapping the coverage provided by antennas using autonomous drones and data fusion techniques. The primary objective of the system is to leverage a contextual layer that provides essential information regarding mobile antennas and their characteristics, such as type, operating frequency, and geodetic coordinates, among other relevant parameters. Based on this contextual layer, the system infers an initial coverage map, which relies entirely on analytical models to estimate free-space power loss as a function of signal propagation distance and frequency. This preliminary estimate is subsequently refined and adjusted to generate a coverage map that more accurately reflects real-world conditions. The refinement process is carried out through measurements acquired at various spatial locations by the drone, which undergo a spatial regression process that employs fully connected neural networks. The study demonstrates how small-scale autonomous drone missions can efficiently sample regions within the target mapping area. These sampled data points enable, through the fusion and filtering process, the spatial propagation of power loss corrections, thereby enhancing the contextual layer and yielding a significantly more precise coverage map.

The reduction of dimensionality in machine learning and artificial intelligence problems constitutes a pivotal element in the simplification of models, significantly enhancing both their performance and execution time. This process enables the generation of results more rapidly while also facilitating the scalability and optimization of systems that rely on such models. Two primary approaches are commonly employed to achieve dimensionality reduction: feature selection-based methods and those grounded in feature extraction. In this paper, we propose a distance-correlation feature space, upon which we define a dimensionality reduction algorithm based on space transformations and graph embeddings. This methodology is applied in the context of dementia diagnosis through learning models, with the overarching objective of optimizing the diagnostic process.

Instance segmentation is an advanced technique in computer vision that focuses on identifying and classifying each individual object in an image at the pixel level . Unlike semantic segmentation , which groups pixels of similar objects without distinguishing between different instances, instance segmentation assigns unique labels to each object, even if they are of the same class. This makes it possible not only to detect the presence and category of objects in an image but also to locate each specific instance and clearly distinguish them from each other. This problem not only advances the technical and theoretical understanding of how machines see and process digital images, but also has a direct impact on various industries and sectors where computer vision is an essential part of the system. In this paper, we present the current deep learning-based technologies, the metrics used for their evaluation, and a review of general and concrete datasets in general and drone-specific contexts. The results of this study provide a compendium of easily deployable deep learning-based technologies. This review paper aims to accelerate the process of understanding and using instance segmentation technologies for the reader.

This paper shows an innovative approach to road monitoring by integrating autonomous drones and sensor fusion within a computer vision-based system. By employing different sets of algorithms, drones equipped with cameras, GPS, and ultrasonic distance sensors can efficiently detect and geolocate road damages, providing crucial data for maintenance and infrastructure management. The system’s key components include automated planning, autonomous flight capabilities, object detection, and sensor fusion techniques, enhancing scalability and adaptability. The main objective is to use context information to compute a flying plan where the subsequent detection of defects allows us to expand and enhance GIS data. The implementation of such a system holds significant potential for improving road safety and optimizing maintenance costs, marking a notable advancement in the convergence of autonomous technologies, sensor fusion, and computer vision for effective road surveillance.

Climbing has gained popularity in recent years and encompasses various disciplines, among which bouldering stands out as one of the most well-known. Determining the difficulty of a bouldering route is a challenging task due to the varying combinations of handholds. Accurate assessment of the difficulty of a route is important, as it allows a climber’s ability to be measured against that of other climbers. Furthermore, climbers must gradually progress from less to more difficult routes to improve their skills effectively. The difficulty rating of a route poses a complex problem, and previous research has explored machine learning techniques, including deep learning, to address it. We propose a novel approach based on a series of transformations in the representation space that incorporates expert knowledge to improve the classification of climbing route difficulty. We introduce a cascade classifier ensemble (CCE) model specifically designed for levelling classification. Compared to similar works, this model also incorporates explainable artificial intelligence (XAI) mechanisms. With the CEE model, the relevance of specific handholds in the transition from one level of difficulty to the next is obtained. Finally, we illustrate how our proposed model not only achieves good results but also empowers climbers to identify the most crucial sections in each level of difficulty. • The transformative pipeline includes expert knowledge, enhancing classifier performance. • CCE excels in classifying climbing route difficulty. • CCE identifies key handholds to level up.

In the field of the intralogistics industry, we present DMZoomNet, a novel architecture that combines deep learning-based detectors with distance information to enhance object detection performance. Evaluation of our approach is conducted using the LOCO dataset, one of the few open source datasets available specifically designed for intralogistics scenarios. By comparing DMZoomNet with existing detectors and object detection methods, we demonstrate its superiority in several object detection metrics within complex intralogistics environments, such as warehouses densely packed with objects. This work contributes to the advancement of object detection techniques in the intralogistics industry and paves the way for future research and applications in this domain.

Within the EUROCONTROL Air Traffic Management (ATM) architecture, the Surveillance Analysis Support System for ATC Centres (SASS-C) is dedicated to the validation of Air Navigation Service Providers. One of its applications is the opportunity Traffic Reconstruction (OTR), which seeks to reconstruct all trajectories by combining noisy measurements from different sensors. It relies on association, tracking and fusion algorithms to determine the true motion of the aircraft on both the vertical and horizontal axes, alongside contextual information useful for identifying the aircraft’s flight mode at any given time. This paper focuses on the filtering of the vertical dimension and outlines certain problems present in the system: the effect of quantized measurements, the appearance of mode of flight transition overshoots and the low reactivity of the filter to abrupt transitions. These problems are analyzed and preliminary improvements according to the ATC context are implemented to overcome them. To demonstrate the improvement, a comparison between the proposed system and the original one is carried out through synthetic trajectory analysis.

UAVs are a cutting-edge technology whose use is currently highly restricted due to their potentially dangerous characteristics, and due to the lack of legislation adopting them and allowing a safe control of these vehicles. Unmanned Air System Traffic Management (UTM) initiatives seek to regularise their use by means of validation and monitoring techniques for the trajectories of these aircraft, both before flight and in real time. For this purpose, in the UTM framework, drones will be collaborative using similar systems to AIS and ADSB for ships or aircraft vehicles. Currently there are no UAV trajectory datasets that allow research in this field, so in this paper a dataset composed of the position and kinematics of the drones over time has been designed. By means of this dataset it is possible to develop and evaluate machine learning methods that help the verification entity to fulfil its functionalities. In this work we propose a first approach to extract useful information by classifying the type of drone based on its movement dynamics. This information would be useful in the identification of the validity of the proposed trajectory for the airframe indicated by the user. The code used for this research is available at https://github.com/DavidSanpedrochez/UAVTrackClassification

Due to Industry 4.0, machines can be connected to their manufacturing processes with the ability to react faster and smarter to changing conditions in a factory. Previously, Internet of Things (IoT) devices could only collect and send data to the cloud for analysis. However, the increasing computing capacity of today’s devices allows them to perform complex computations on-device, resulting in edge computing. Edge devices are a fundamental component of modern, distributed real-world artificial intelligence (AI) systems in Industry 4.0 environments. As a result, edge computing extends cloud computing capabilities by bringing services near the edge of a network and thus supports a new variety of AI services and machine learning (ML) applications. However, there is a large difference between designing and training an ML model, potentially in the cloud, to create ML services that can be deployed and consumed on the edge. This article presents an ML workflow based on ML operations (MLOps) over the Thinger.io IoT platform to streamline the transition from model training to model deployment on edge devices. The proposed workflow is composed of different elements, such as the ML training pipeline, ML deployment pipeline, and ML workspace. Similarly, this article describes the ease of design and deployment of the proposed solution in a real environment, where an anomaly detection service is implemented for detecting outliers on temperature and humidity measurements. The performance tests performed over the ML pipeline steps and the ML service throughput on the edge indicate that this workflow adds minimum overhead to the process, providing a more reliable, reusable, and productive environment.

Abstract The process of parametric equalization of musical pieces seeks to highlight their qualities by cutting and/or stimulating certain frequencies. In this work, we present a neural model capable of equalizing a song according to the musical genre that is being played at a given moment. It is normal that (1) the equalization should adapt throughout the song and not always be the same for the whole song; and (2) songs do not always belong to a specific musical genre and may contain touches of different musical genres. The neural model designed in this work, called CONEqNet (convolutional music equalizer network), takes these aspects into account and proposes a neural model capable of adapting to the different changes that occur throughout a song and with the possibility of mixing nuances of different musical genres. For the training of this model, the well-known GTzan dataset, which provides 1,000 fragments of songs of 30 seconds each, divided into 10 genres, was used. The paper will show proofs of concept of the performance of the neural model.

Abstract According to the World Health Organization, between and , the proportion of the world's population over will double, from to . In absolute numbers, this age group will increase from million to billion in the course of half a century. It is a reality that most of them prefer to live alone, so it is necessary to look for mechanisms and tools that will help them to improve their autonomy. Although in recent years, we have been living in a veritable explosion of domotic systems that facilitate people's daily lives, it is also true that there are not many tools specifically aimed at this sector of the population. The aim of this paper is to present a potential solution to the monitoring of activity of daily living in the least intrusive way for people. In this case, anomalous patterns of daily activities will be detected by analysing the daily consumption of household appliances. People who live alone usually have a pattern of daily behaviour in the use of household appliances (coffee machine, microwave, television, etc.). A neuronal model is proposed for the detection of abnormal behaviour based on an autoencoder architecture. This solution will be compared with a variational autoencoder to analyse the improvements that can be obtained. The well‐known dataset called UK‐DALE will be used to validate the proposal.

Abstract Conversational systems have become an element of everyday life for billions of users who use speech‐based interfaces to services, engage with personal digital assistants on smartphones, social media chatbots, or smart speakers. One of the most complex tasks in the development of these systems is to design the dialogue model, the logic that provided a user input selects the next answer. The dialogue model must also consider mechanisms to adapt the response of the system and the interaction style according to different groups and user profiles. Rule‐based systems are difficult to adapt to phenomena that were not taken into consideration at design‐time. However, many of the systems that are commercially available are based on rules, and so are the most widespread tools for the development of chatbots and speech interfaces. In this article, we present a proposal to: (a) automatically generate the dialogue rules from a dialogue corpus through the use of evolving algorithms, (b) adapt the rules according to the detected user intention. We have evaluated our proposal with several conversational systems of different application domains, from which our approach provided an efficient way for adapting a set of dialogue rules considering user utterance clusters.

Abstract As life expectancy increases, it has become more necessary to find ways to support healthy ageing. A number of active ageing initiatives are being developed nowadays to foster healthy habits in the population. This paper presents our contribution to these initiatives in the form of a multimodal conversational coach that acts as a coach for physical activities. The agent can be developed as an Android app running on smartphones and coupled with cheap widely available sport sensors in order to provide meaningful coaching. It can be employed to prepare exercise sessions, provide feedback during the sessions, and discuss the results after the exercise. It incorporates an affective component that informs dynamic user models to produce adaptive interaction strategies.

This research explores the ability to extract knowledge about the associations among agricultural products which allows to improve the prediction of future consumption in the local markets of the Andean region of Ecuador. This commercial activity is carried out using Alternative Marketing Circuits (CIALCO), seeking to establish a direct relationship between producer and consumer prices, and promote buying and selling among family groups. The fusion of information from spatially located heterogeneous data sources allows to establish the best association rules between data sources (several products in several local markets) to infer a significant improvement in spatial prediction accuracy for sales future agricultural products.

Abstract Smart mobile devices have fostered new learning scenarios that demand sophisticated interfaces. Multimodal conversational agents have became a strong alternative to develop human‐machine interfaces that provide a more engaging and human‐like relationship between students and the system. The main developers of operating systems for such devices have provided application programming interfaces for developers to implement their own applications, including different solutions for developing graphical interfaces, sensor control and voice interaction. Despite the usefulness of such resources, there are no strategies defined for coupling the multimodal interface with the possibilities that these devices offer to enhance mobile educative apps with intelligent communicative capabilities and adaptation to the user needs. In this paper, we present a practical m‐learning application that integrates features of Android application programming interfaces on a modular architecture that emphasizes interaction management and context‐awareness to foster user‐adaptively, robustness and maintainability.

Abstract Employing multiple camera viewpoints in the recognition of human actions increases performance. This paper presents a feature fusion approach to efficiently combine 2D observations extracted from different camera viewpoints. Multiple‐view dimensionality reduction is employed to learn a common parameterization of 2D action descriptors computed for each one of the available viewpoints. Canonical correlation analysis and their variants are employed to obtain such parameterizations. A sparse sequence classifier based on L1 regularization is proposed to avoid the problem of having to choose the proper number of dimensions of the common parameterization. The proposed system is employed in the classification of the Inria Xmas Motion Acquisition Sequences (IXMAS) data set with successful results.

The paper presents a group behavior representation and an application in the 2 vs. 2 basketball domain. Furthermore a set of forty features have been made from the raw information provided by the INEF12 Basketball Dataset. Moreover, from all these features we propose a selection using an algorithm to choose the best features to classify and predict the group behavior. The entire experimental test carried out with Hidden Markov Models algorithms could validate the proposed representation and features selection, in group behavior recognition and 2 vs. 2 basketball specific domain.

The multi-objective neural EDA (MONEDA) was proposed with the aim of overcoming some difficulties of current MOEDAs. MONEDA has been shown to yield relevant results when confronted with complex problems. Furthermore, its performance has been shown to adequately adapt to problems with many objectives. Nevertheless, one key issue remains to be studied: MONEDA scalability with regard to the number of decision variables. In this paper has a two-fold purpose. On one hand we propose a modification of MONEDA that incorporates an indicator-based selection mechanism based on the HypE algorithm, while, on the other, we assess the indicator-based MONEDA when solving some complex two-objective problems, in particular problems UF1 to UF7 of the CEC 2009 MOP competition, configured with a progressively-increasing number of decision variables.

Representation and reasoning with spatial properties is essential in several application domains where ontologies are being successfully applied; e.g., Information Fusion systems. This requires a full characterization of the semantics of relations such as adjacent, included, overlapping, etc. Nevertheless, ontologies are not expressive enough to directly support widely-use spatial or topological theories, such as the Region Connection Calculus (RCC). In addition, these properties must be properly instantiated in the ontology, which may require expensive calculations. This paper presents a practical approach to represent and reason with topological properties in ontology-based systems, as well as some optimization techniques that have been applied in a video-based Information Fusion application.

Abstract: Current Computer Vision systems are expected to allow for the management of data acquired by physically distributed cameras. This is especially the case for modern surveillance systems, which require communication between components and a combination of their outputs in order to obtain a complete view of the scene. Information fusion techniques have been successfully applied in this area, but several problems remain unsolved. One of them is the increasing need for coordination and cooperation between independent and heterogeneous cameras. A solution to achieve an understanding between them is to use a common and well‐defined message content vocabulary. In this research work, we present a formal ontology aimed at the symbolic representation of visual data, mainly detected tracks corresponding to real‐world moving objects. Such an ontological representation provides support for spontaneous communication and component interoperability, increases system scalability and facilitates the development of high‐level fusion procedures. The ontology is used by the agents of Cooperative Surveillance Multi‐Agent System, our multi‐agent framework for multi‐camera surveillance systems.

Soft computing methods, and Multi-Objective Evolutionary Algorithms (MOEAs) in particular, lack a general convergence criterion which prevents these algorithms from detecting the generation where further evolution will provide little improvements (or none at all) over the current solution, making them waste computational resources. This paper presents the Least Squares Stopping Criterion (LSSC), an easily configurable and implementable, robust and efficient stopping criterion, based on simple statistical parameters and residue analysis, which tries to introduce as few setup parameters as possible, being them always related to the MOEAs research field rather than the techniques applied by the criterion.

In this work we present a novel progress indicator, called fitness homogeneity indicator (FHI). This indicator improves the other previously discussed indicators as it takes into account all possible processes taking place in the population while not requiring an intensive computation as it relies on the fitness values calculated for the individuals. It is also capable of equally detecting success and failure scenarios, hopefully making an early detection of the second case.

In this work we analyze the model-building issue and the requirements it imposes on the learning paradigm being used. We argue that error-based learning, the class of learning most commonly used in MOEDAs, is responsible for current MOEDA underachievement. We present ART as a viable alternative and present a novel algorithm called multi-objective ART-based EDA (MARTEDA) that uses a Gaussian ART neural network for model-building and an hypervolume based selector as described for the HypE algorithm. We experimentally show that thanks to MARTEDA's novel model-building approach and an indicator-based population ranking the algorithm it is able to outperform similar MOEDAs and MOEAs.

This paper presents a method based on feature selection to obtain sets of human activity recognizers of different complexity. Classifiers for human activity recognition are built exploring a space of candidate feature subsets, trying to maximize the accuracy of a classifier trained with them. At the same time, the size of the selected feature subset is minimized. The accuracy of a classifier tends to grow with the number of features, but in a real time task, like human activity recognition, the number of features used has to be minimized, because its growing involves a slower processing rate. A set of solutions with different trade-offs between accuracy and number of features may be achieved modeling the problem of feature selection using multiobjective optimization (MO), where both measures are optimized at the same time. To solve the MO problem, multiobjective optimization evolutionary algorithms (MOEA) are going to be used. MOEA methods based on Pareto dominance not only find an optimal solution for the problem, they find a set of different optimal solutions so called Pareto-optimal set. A set of activity recognizers of different complexities is found using this approach. Having a set of different solutions allows the designer to choose the one that best fits its requirements. The method will be applied using a hidden Markov model as classifier. Results of the use of the method for recognizing different instantaneous human activities are discussed.

Classical tracking methods are often insufficient when dealing with complex scenarios. In order to solve tracking errors, innovative techniques based on the use of information about the context of the scene have been proposed. Context information ranges from precise measures computed on the pixels of the object neighborhood to high level representations of the entities and the activities of the scene. In this work, we focus on the second approach and propose an ontology-based extension of a general tracking procedure that reasons with abstract context descriptions to improve its accuracy. We describe the design of this extension and how reasoning is performed, as well as its advantages in surveillance scenarios.

In this work we put forward a comprehensive study on the design of global stopping criteria for multi-objective optimization. We describe a novel stopping criterion, denominated MGBM criterion that combines the mutual domination rate (MDR) improvement indicator with a simplified Kalman filter that is used for evidence gathering process. The MDR indicator, which is introduced along, is a special purpose solution meant for the stopping task. It is capable of gauging the progress of the optimization with a low computational cost and therefore suitable for solving complex or many-objective problems. The viability of the proposal is established by comparing it with some other possible alternatives. It should be noted that, although the criteria discussed here are meant for MOPs and MOEAs, they could be easily adapted to other softcomputing or numerical methods by substituting the local improvement metric with a suitable one.

This paper deals with the multi-objective definition of video compression and it solving using the NSGA-II algorithm. We define the video compression as a problem including two competing objectives and we try to find a set of near-optimal solutions so called Pareto-optimal solutions instead of a single optimal solution. This will be applied to a new codec that is patent pending, which needs some optimizations before it release. The compression is achieved over a standard video, commonly used for video performance measurement. Also we present the NSGA-II convergence speed and discuss the suitability of MOEAs in this scope.

The need for a stopping criterion in MOEA's is a repeatedly mentioned matter in the domain of MOOP's, even though it is usually left aside as secondary, while stopping criteria are still usually based on an a-priori chosen number of maximum iterations. In this paper we want to present a stopping criterion for MOEA's based on three different indicators already present in the community. These indicators, some of which were originally designed for solution quality measuring (as a function of the distance to the optimal Pareto front), will be processed so they can be applied as part of a global criterion, based on estimation theory to achieve a cumulative evidence measure to be used in the stopping decision (by means of a Kalman filter). The implications of this cumulative evidence are analyzed, to get a problem and algorithm independent stopping criterion (for each individual indicator). Finally, the stopping criterion is presented from a data fusion perspective, using the different individual indicators' stopping criteria together, in order to get a final global stopping criterion.

The multi-objective optimization neural estimation of distribution algorithm (MONEDA) was devised with the purpose of dealing with the model-building issues of MOEDAs and, therefore address their scalability.

Trust and reputation management play an important role in agent-based recommender systems. Although several protocols and ontologies of agents using trust and reputation has been proposed, none of them has been so extensively used and implicitly accepted by research community as those from agent reputation and trust (ART in advance) testbed. The motivation of this adaptation is to facilitate the use of ART principles in real distributed applications instead of a centralized testbed for experimentation. This paper presents an adaptation of the protocols proposed by ART testbed to a codification for the most popular agent platform: JADE. This implementation follows a coherent API with the FIPA protocols included in JADE distribution for an easy use. We also complement the behaviours of corresponding initiators and responders of the protocols with an ontology formed by a collection of concepts, predicates and agent actions that may represent as the ART application domain as any other service-oriented domain. The proposal has been designed to be applied in domains where multi-agent e-commerce solutions are needed. Future work includes the integration of this ontology and protocols in context-aware scenarios such as an airport.

This paper addresses a real-world engineering design requiring the application of effective and global optimization techniques. The problem it deals with is the design of nonlinear tracking filters under up to several hundreds of performance specifications. The suitability of different evolutionary computation techniques for solving multiobjective problems is explored, contrasting the performance achieved with recent multiobjective evolutionary algorithm (MOEAs) proposals and different aggregation schemes. In particular, a new scheme is proposed to build a fitness function based on an operator that selects worst cases of multiple specifications in different situations. They have been evaluated in the design of an air traffic control (ATC) tracking filter that should accomplish a specific normative with 264 specifications. Results show their performance in terms of effectiveness and computational load, comparing their capability to scale the problem with respect to problem size.

In order to comprehend the advantages and short-comings of each model-building algorithm they should be tested under similar conditions and isolated from the MOEDA it takes part of. In this work we will assess some of the main machine learning algorithms used or suitable for model-building in a controlled environment and under equal conditions. They are analyzed in terms of solution accuracy and computational complexity. To the best of our knowledge a study like this has not been put forward before and it is essential for the understanding of the nature of the model-building problem of MOEDAs and how they should be improved to achieve a quantum leap in their problem solving capacity.

In this work the data association problem in visual tracking is formulated as a combinatorial hypotheses search with a heuristic evaluation function taking into account structural and specific information such as distance, shape, colour, etc. In order to guarantee real time performance, the search process has a time limit to explore alternative solutions. This time defines the upper bound of the number of evaluations depending on the efficiency of the search algorithm. Estimation distribution algorithms (EDA) is proposed as an efficient evolutionary computation technique to search in this hypothesis space. Then, an exhaustive comparison of the performance of alternative algorithms is carried out considering complex representative situations in real video sets.

In this paper we explore the model-building issue of multiobjective optimization estimation of distribution algorithms. We argue that model-building has some characteristics that differentiate it from other machine learning tasks. A novel algorithm called multiobjective neural estimation of distribution algorithm (MONEDA) is proposed to meet those characteristics. This algorithm uses a custom version of the growing neural gas (GNG) network specially meant for the model-building task. As part of this work, MONEDA is assessed with regard to other classical and state-of-the-art evolutionary multiobjective optimizers when solving some community accepted test problems.

This paper investigates combinatorial and probabilistic approaches to real-time video target tracking. Of special interest are real-world scenarios, in which the presence of multiple targets and complex background pose a non-trivial challenge to automated trackers. Object tracking in an exemplary surveillance video sequence is accomplished by means of selected visual tracking techniques, based on two families of methods, combinatorial data association and Particle Filters. Based on the detailed analysis of the performance of the trackers tested, the advantages, complementary failure modes and computational requirements of each method have been identified. Taking into account the results obtained, the hybrid strategy for improved tracking performance is suggested, bringing together the best complementary features of the different tracking methods.

In this paper an approach for multi-sensor coordination in a multiagent visual sensor network is presented. A belief-desire-intention model of multiagent systems is employed. In this multiagent system, the interactions between several surveillance-sensor agents and their respective fusion agent are discussed. The surveillance process is improved using a bottom-up/top-down coordination approach, in which a fusion agent controls the coordination process. In the bottom-up phase the information is sent to the fusion agent. On the other hand, in the top-down stage, feedback messages are sent to those surveillance-sensor agents that are performing an inconsistency tracking process with regard to the global fused tracking process. This feedback information allows to the surveillance-sensor agent to correct its tracking process. Finally, preliminary experiments with the PETS 2006 database are presented.

A surveillance system that fuses data from several data sources is more robust than those which depends on a single source of input. Fusing the information acquired by a vision system is a difficult task since the system needs to use reliable models for errors and take into account bad performance when taking measurements. In this research, we use a bidimensional object correspondence and tracking method based on the ground plane projection of the blob centroid. We propose a robust method that employs a two phase algorithm which uses a heuristic value and context information to automatically combine each source of information. The fusion process is carried out by a fusion agent in a multi-agent surveillance system. The experimental results on real video sequences have showed the effectiveness and robustness of the system.

This work addresses off-line accurate trajectory reconstruction for air traffic control. We propose the use of specific dynamic models after identification of regular motion patterns. Datasets recorded from opportunity traffic are first segmented in motion segments, based on the mode probabilities of an IMM filter. Then, reconstruction is applied with an optimal smoothing filter operating forward and backward. The parameters describing the specific modes are estimated and then used as external input for smoothing filters. The performance of this approach is compared with a method based on interpolation B-splines. Comparative results on simulated and real data are discussed at the end.

Video target tracking is a complex task, specially when the tracking system is expected to work well in different scenarios. For this reason, this paper proposes an architecture based on a two layer image-processing modules: general tracking layer (GTL) and context layer (CL). GTL describes a generic multipurpose tracking process for video surveillance systems. CL is designed as a symbolic reasoning system that manages the symbolic interface data between GTL modules in order to assess a specific situation and take the appropriate decision. CL intervenes at three different stages of the tracking process, these are initialization, association and update. Our architecture has been tested in two different scenarios to show the advantages in improved performance and output continuity.

In this work we present a novel and efficient algorithm independent stopping criterion, called the MGBM criterion,suitable for Multi-objective Optimization Evolutionary Algorithms(MOEAs).The criterion, after each iteration of the optimization algorithm, gathers evidence of the improvement of the solutions obtained so far. A global (execution wise) evidence accumulation process inspired by recursive Bayesian estimation decides when the optimization should be stopped. Evidence is collected using a novel relative improvement measure constructed on top of the Pareto dominance relations. The evidence gathered after each iteration is accumulated and updated following a rule based on a simplified version of a discrete Kalman filter.Our criterion is particularly useful in complex and/or high-dimensional problems where the traditional procedure of stopping after a predefined amount of iterations cannot beused and the waste of computational resources can induceto a detriment of the quality of the results.Although the criterion discussed here is meant for MOEAs,it can be easily adapted to other soft computing or numerical methods by substituting the local improvement metric witha suitable one.

The growing interest in surveillance in public, military and commercial scenarios is increasing the need to create intelligent or automated distributed visual surveillance systems. Many applications based on distributed resources use the software agent paradigm. In this work, a multi-agent framework is applied to coordinate an indoor-surveillance system based on video cameras. The capacity of coordination will allow the improvement of the global image and the effectiveness of the task distribution. Software agents are embedded in each camera and control the capture parameters. The multi-agent framework allows the coordination of the acquisition procedure based on high level messages and the fusion of information among agents. The agent paradigm uses the internal interpretation of the situation from each agent to improve the global coordination

This work addresses the application of a machine-learning approach to classify ATC trajectory segments from recorded opportunity traffic. It is based on the mode probabilities estimated by an IMM tracking filter operating forward and backward over available data. A learning algorithm creates a rule base for classification from these data, once they have been properly prepared. Performance of this data-driven classification system is compared with a more conventional approach based on transition detection on simulated and real data of representative situations. The offline processing of real data allows an accurate classification of manoeuvring segments, with the possibility of synthesizing ground truth lines for performance evaluation

Many methods to codify artificial neural networks have been developed to avoid the disadvantages of direct encoding schema, improving the search into the solution's space. A method to analyse how the search space is covered and how are the movements along search process applying genetic operators is needed in order to evaluate the different encoding strategies for multilayer perceptrons (MLP). In this paper, the generative capacity, this is how the search space is covered for a indirect scheme based on cellular systems, is studied. The capacity of the methods to cover the search space (topologies of MLP space) is compared with the direct encoding scheme.

Open electronic communities may bring together people geographically and culturally unrelated to each other. In this context, taking costly decisions depends on the expectations created according to past behaviour of others. This kind of information is usually called reputation and it is one of the most significant factors to trust merchants and recommenders in electronic commerce interactions. When agents are acting on behalf of humans in such commercial scenarios, they should represent and reason about trust and reputation as humans do. In this paper a trust management mechanism tackles the vague, subjective and uncertain information about others using fuzzy sets. The operations defined over such fuzzy sets updates the reputation of merchants according to the general situation faced. This trust management mechanism is applied to a multiagent system of merchants, recommenders and buyers, where collaborative recommendations coexist with competitive intentions. The developed multi-agent system is used to compare the level of success of predictions obtained from the fuzzy computations with some of the most well known (crisp) reputation mechanisms: ebay, bizrate, sporas and regret when the behaviour of merchants change in different degrees. Finally, the potential benefits of using fuzzy sets to manage reputation in multi-agent systems are analyzed according to the excellent experimental results shown.

The use of distributed artificial intelligence (DAI) techniques, particularly the multiagent systems theory, in a decentralized architecture, is proposed to manage cooperatively, all sensor tasks in a network of (air) surveillance radars with capabilities for autonomous operation. At the multisensor data fusion (DF) center, the fusion agent will periodically deliver to sensor agents a list with the system-level tasks that need to be fulfilled. For each system task, indications about its system-level priority are included (inferred global necessity of fulfilling the task) as well as the performance objectives that are required, expressed in different terms depending on the type of task (sector surveillance, target tracking, target identification, etc.). Periodically, the local manager at each sensor (the sensor agent) will decide on the list of sensor-level tasks to be executed by its sensor, providing also the sensor-level priority and performance objectives for each task. The problem of sensor(s)-to-task(s) assignment (including decomposition of system-level tasks into sensor-level tasks and translation of system-level performance requirements to sensor-level performance objectives) is the result of a negotiation process performed among sensor agents, initiated with the information sent to them by the fusion agent. With types of agents, a symbolic bottom-up fuzzy reasoning process is performed that considers the available fused or local target tracks, surveillance sectors data, and (external) intelligence information. As a result of these reasoning processes, performed at each agent planning level, the priorities of system-level and sensor-level tasks will be inferred and applied during the negation process. © 2003 Wiley Periodicals, Inc.

This paper describes the optimisation of some parameters of an optical character recognition system (OCR). The optimisation is performed by means of evolution strategies (ES) in order to maximize the pattern discrimination. The pattern set is a vectorial representation of the character set. The OCR is applied to identify the tail number of an aircraft moving on the airfield runway. The proposed approach is discussed together with some results obtained on a benchmark data set of aircraft tail numbers.

Designing the optimal neural net (NN) architecture can be formulated as a search problem in the architectures space, where each point represents an architecture. The search space of all possible architectures is very large, and the task of finding the simplest architecture may be an arduous and mostly a random task. Methods based on indirect encoding have been used to reduce the chromosome length. In this paper, a new indirect encoding method is proposed and an analysis of the generative capacity of the method is presented.

A new approach for data association problems in video image sequences is presented, which uses JPDA formulation adapted to cope with video data peculiarities. A correlation level is computed to weight each blob contribution to each track, by means of a fuzzy system integrating different heuristics inferred from system performance under real situations. Results obtained in representative ground operations show the system capabilities to solve complex scenarios and improve tracking accuracy.

We present a design procedure, based on ES optimisation, to parametrize IMM tracking structures for air traffic control applications. The objective, not addressed in the available bibliography on IMM filters, is to find the most suitable parameters for a selected IMM structure, accordingly to a determined a set of performance specifications. Two alternative structures are analysed following this procedure, showing the best performance achievable by each one, in terms of the proximity of the performance metrics to the specified values.

Solving problems in dynamic and heterogeneous environments where information sources change their format representation and stored data is very complex. In previous work we presented a system called MAPWeb (Multiagent Planning on the Web) that tried to solve these problems by integrating artificial intelligence planning techniques within the multiagent framework. Basically, MAPWeb allows cooperative work between planning agents and Web agents. The purpose of MAPWeb is to find solutions to travel problems. In order to give detailed solutions, MAPWeb uses information gathering techniques to retrieve travel information that is made available by many different companies. However, Web access to the information sources is quite time expensive. In this paper, we try to minimize the number of Web queries by using caching techniques based on relational databases. Experimental results show that the reduction in Web access time is quite important, while maintaining the number of solutions found.

It is possible to find any kind of useful information in the Web. However, there are serious problems in retrieving, managing and using this information, due to its vastness. Different approaches have been developed to avoid those problems (search engines, metasearch engines, spiders, softbots, intelligent agents or Web agents). This paper is based on one of these systems which uses a set of heterogeneous intelligent software agents to achieve these previous tasks. Two different agents compose the system: Web agents developed to retrieve information from a specific Web source and meta Web agents developed to select the appropriated Web agent to search the necessary information. Each Web agent retrieves, filters and stores the information from the Web to improve system performance. The meta Web agents need to represent and classify the behavior of different Web agents. In this work a fuzzy system that helps to classify the behavior of the Web agents is presented. The meta Web agent calculates the appropriateness of existing agent behavior, using different distances that are analyzed in the paper. The behavior classification is used to decide which Web agent is requested for information by the meta Web agent.

Agents acting on behalf of human users should cooperate with others and reason about their expected behavior in order to avoid deceptions and frauds. This knowledge about others will be used as a means to judge their reputation, and it involves how the services were provided, and whether they suited the particular expectations of the human user represented by the agent. This paper outlines an application of the most popular (and theoretically sound) agent architecture to that problem. This is the so called BDI architecture. Our research describes the beliefs, desires, intentions, and the relationships among them relevant to the given dominion of an agent reasoning about reputation. Due to the subjective nature of such reputations, we use fuzzy sets to represent them.

This paper purposes a new computation model that integrates some of the most important agent technology concepts, which are based on the agent concept as the keyword from which to establish the three main pillars of the paradigm. First, a new kind of internal and external structure of the agent and of the environment where it resides based on dynamical components is described. Second, the application of a new variety of evolutionary heuristic existing at different levels of abstraction, which is based on concepts obtained from the evolutionary computation is described. And finally, new software engineering techniques are introduced to manage the design and development of these systems.

A new coevolutive method, called Uniform Coevolution, is introduced, to learn weights for a neural network controller in autonomous robots. An evolutionary strategy is used to learn high-performance reactive behavior for navigation and collision avoidance. The coevolutive method allows the evolution of the environment, to learn a general behavior able to solve the problem in different environments. Using a traditional evolutionary strategy method without coevolution, the learning process obtains a specialized behavior. All the behaviors obtained, with or without coevolution have been tested in a set of environments and the capability for generalization has been shown for each learned behavior. A simulator based on the mini-robot Khepera has been used to learn each behavior. The results show that Uniform Coevolution obtains better generalized solutions to example-based problems.

The class of feedforward neural networks trained with back-propagation admits a large variety of specific architectures applicable to approximation pattern tasks. Unfortunately, the architecture design is still a human expert job. In recent years, the interest to develop automatic methods to determine the architecture of the feedforward neural network has increased, most of them based on the evolutionary computation paradigm. From this approach, some perspectives can be considered: at one extreme, every connection and node of architecture can be specified in the chromosome representation using binary bits. This kind of representation scheme is called the direct encoding scheme. In order to reduce the length of the genotype and the search space, and to make the problem more scalable, indirect encoding schemes have been introduced. An indirect scheme under a constructive algorithm, on the other hand, starts with a minimal architecture and new levels, neurons and connections are added, step by step, via some sets of rules. The rules and/or some initial conditions are codified into a chromosome of a genetic algorithm. In this work, two indirect constructive encoding schemes based on grammars and cellular automata, respectively, are proposed to find the optimal architecture of a feedforward neural network.

Neural networks (NN) can be used as controllers in autonomous robots. The specific features of the navigation problem in robotics make generation of good training sets for the NN difficult. An evolution strategy (ES) is introduced to learn the weights of the NN instead of the learning method of the network. The ES is used to learn high performance reactive behavior for navigation and collision avoidance. No subjective information about "how to accomplish the task" has been included in the fitness function. The learned behaviors are able to solve the problem in different environments; therefore, the learning process has the proven ability to obtain a specialized behavior. All the behaviors obtained have been tested in a set of environments and the capability of generalization is shown for each learned behavior. A simulator based on the mini-robot, Khepera, has been used to learn each behavior.

One of the major problems related to classifier systems is the loss of rules. This loss is caused by the genetic algorithm being applied on the entire population of rules jointly. Obviously, the genetic operators discriminate rules by the strength value, such that evolution favours the generation of the stronger rules. When the learning system works in an environment in which it is possible to generate a complete training set, the strength of the rules of the CS will reflect the relative relationship between rules satisfactorily and, therefore, the application of the genetic algorithm will produce the desired effects. However, when the learning process presents individual cases and allows the system to learn gradually from these cases, each learning interval with a set of individual cases can lead the strength to be distributed in favour of a given type of rules that would in turn be favoured by the genetic algorithm. Basically, the idea is to divide rules into groups such that they are forced to remain in the system. This contribution is a method of learning that allows similar knowledge to be grouped. A field in which knowledge-based systems researchers have done a lot of work is concept classification and the relationships that are established between these concepts in the stage of knowledge conceptualization for later formalization. This job of classifying and searching relationships is performed in the proposed classifier systems by means of a mechanism. Tags, that allows the classification and the relationships to be discovered without the need for expert knowledge.

The navigation problem involves how to reach a goal avoiding obstacles in dynamic environments. This problem can be faced considering reactions and/or sequences of actions. Classifier Systems (CS) have proven their ability of continuous learning, however they have some problems in reactive systems. A modified CS is proposed to overcome these problems. Two special mechanisms are included in the developed CS to allow the learning of both reactions and sequences of actions. This learning process involves two main tasks: first, discriminating between rules and second, the discovery of new rules to obtain a successful operation in dynamic environments. Different experiments have been carried out using a mini-robot Khepera to find a generalized solution. The results show the ability of the system for continuous learning and adaptation to new situations.