[Show abstract][Hide abstract] ABSTRACT: The use of interest point detectors and SIFT descriptors for face recognition is studied in this paper. There are two main novelties with respect to previous approaches using SIFT features. First, the use of two scale-invariant interest point detectors (namely, Harris-Laplace and difference of Gaussians) which are combined in order to detect both corner-like structures and blob-like structures in face images. Second, the distance measure used, which takes into account both the number of matching points found between two images (according to their SIFT descriptors) and the coherence of these matches in terms of scales, orientations and spacial configuration. The results obtained with our model-based algorithm are compared with those of a classic appearance-based face recognition method (PCA) over two different face databases: the well-known AT&T database and a face database created at our university.
Automatic Face & Gesture Recognition, 2008. FG '08. 8th IEEE International Conference on; 10/2008
[Show abstract][Hide abstract] ABSTRACT: An appearance based object detection system is presented, whose main goal is the detection of free-form objects in non structured environments. A multichannel and multiscale approach is used. Two different feature extraction methods are combined to generate a model for each object: Principal Component Analysis (PCA) and Non-negative Matrix Factorization with Sparsity Constraints (NMFSC). An object-dependant adjustment of channel weighting allows to detect different objects with high accuracy. The detection threshold is also automatically adjusted depending on the object to be detected.
The Seventh IASTED International Conference on Visualization, Imaging and Image Processing; 08/2007
[Show abstract][Hide abstract] ABSTRACT: In this paper, a teaching application for remote real-time execution of physical process controllers is presented. This application has been developed using the platform Matlab/Simulink. The motivation of this work is based on the little availability of real physical systems or laboratories to perform the experiments in control courses. In this way, control lab assignments with various physical processes present in the remote laboratories can be performed. Also, some examples that show the validity and applicability of the presented architecture are introduced.
Industrial Electronics, 2007. ISIE 2007. IEEE International Symposium on; 07/2007
[Show abstract][Hide abstract] ABSTRACT: In this paper we describe an approach that builds three dimensional maps using visual landmarks extracted from images of an unmodified environment. We propose a solution to the Simultaneous Localization and Mapping (SLAM) problem for autonomous mobile robots using visual landmarks. Our map is represented by a set of three dimensional landmarks referred to a global reference frame, each landmark contains a visual descriptor that partially differentiates it from others. Significant points extracted from stereo images are used as natural landmarks, in particular we employ SIFT features found in the environment. We estimate both the map and the path of the robot using a Rao-Blackwellized particle filter, thus the problem is decomposed into two parts: one estimation over robot paths using a particle filter, and N independent estimations over landmark positions, each one conditioned on the path estimate. We actively track visual landmarks at a local neighbourhood and select only those that are more stable. When a visual feature has been observed from a significant number of frames it is then integrated in the filter. By this procedure, the total number of landmarks in the map is reduced, compared to prior approaches. Due to the tracking of each landmark, we obtain different examples that represent the same natural landmark. We use this fact to improve data association. Finally, efficient resampling techniques have been applied, which reduces the number of particles needed and avoids the particle depletion problem.
ICINCO 2006, Proceedings of the Third International Conference on Informatics in Control, Automation and Robotics, Robotics and Automation, Setúbal, Portugal, August 1-5, 2006; 01/2006
[Show abstract][Hide abstract] ABSTRACT: A redundancy resolution technique devoted to grasp synthesis is presented. Given a set of contact points and a certain robot arm and gripper, the goal is to select both the best assignment of gripper fingers to contact points and the best joint values that allow the fingers to reach such contact points. The system proposed is based on the generation of an inverse kinematics tree where fast searches can be performed in order to find the optimum configuration. Optimality is defined as similarity to previously stored examples over a hierarchical structure of configuration data, which includes finger assignments and robot joints.
Robotics and Automation, 2005. ICRA 2005. Proceedings of the 2005 IEEE International Conference on; 05/2005
[Show abstract][Hide abstract] ABSTRACT: This work presents a method for identifying real three-dimensional objects in non-controlled backgrounds using independent component analysis to eliminate redundant image information present in each object image. The proposed method is a two-step process that allows a coarse color-based detection and an exact localization using shape information. The paper describes an efficient implementation, making this approach suitable for real-time applications.
[Show abstract][Hide abstract] ABSTRACT: An algorithm for allocating and scheduling tasks in multiprocessor environments is presented. Its main characteristic is its orientation towards machine vision applications. In this sense it deals with the peculiarities of systems which combine generic-type processors with Image Acquisition and Processing Boards. The main goal of the algorithm is total processing time reduction; such are the requirements when we deal with automated industrial inspection applications. By simultaneously tackling the phases of allocation and scheduling, the results obtained are better than those offered by traditional algorithms. The system is applied to a process of citrus fruit inspection, and its performances are also evaluated over randomly generated task graphs.
Parallel Processing and Applied Mathematics, 4th International Conference, PPAM 2001 Naleczow, Poland, September 9-12, 2001, Revised Papers; 01/2001
[Show abstract][Hide abstract] ABSTRACT: Avda. del Ferrocarril s/n 03202 Elche (Alicante) España email@example.com Resumen El presente artículo propone un m étodo de aprendizaje para sistemas teleoperados aplicado a los procesos de agarre. Dado un objeto de una forma arbitraria, los puntos de agarre deben ser calculados automáticamente a partir de los ejemplos proporcionados por el usuario mediante teleoperación. Trabajos previos calculan los puntos de agarre analíticamente a partir de la geometría del objeto. La presente propuesta ofrece mejores resultados dado que considera otros aspectos aparte de la geometría del objeto. Esta información extra está implícita en los ejemplos de aprendizaje proporcionados por el usuario. Se utilizan pinzas paralelas de 2 o 3 dedos, de modo que es necesario calcular un par o un trío de puntos de agarre. El proceso de aprendizaje se realiza en dos etapas mediante árboles de decisión: en la primera etapa se genera un conjunto de puntos válidos entre todos los puntos de contorno; y en la segunda se computan todos los pares o tríos que es posible generar utilizando los puntos v álidos seleccionados previamente. Entre estos pares o tríos se selecciona el agarre óptimo. Los resultados mostrados al final del artículo permiten comprobar la validez de la propuesta. Palabras Clave : Robótica, Teleoperación, Agarre, Aprendizaje automático, Árboles de decisión. 1 APRENDIZAJE EN SISTEMAS TELEOPERADOS Los sistemas teleoperados constituyen un escenario idóneo para la aplicación de técnicas de aprendizaje automático . Durante la operación manual, el usuario comanda los movimientos de un brazo robótico cuyas trayectorias pueden ser archivadas y analizadas. Simultáneamente, el sistema sensorial, normalmente basado en visión artificial, es capaz también de almacenar toda la información de la escena durante la realización de la tarea. La combinación de estos datos (trayectorias y escena) es la información de partida para el algoritmo de aprendizaje; y el resultado es la inferencia de las normas de comportamiento utilizadas por el operario durante la realización de la tarea. Tras varias repeticiones de la misma tarea en modo manual, el robot aprenderá a llevar a cabo la tarea autónomamente. Debe ser tenido en cuenta que este aprendizaje va más allá de la mera repetición de trayectorias; el objetivo es lograr un comportamiento autónomo incluso en entornos cambiantes. En la literatura se encuentran numerosos ejemplos de sistemas basados en aprendizaje. En algunos casos el objetivo es el control a bajo nivel de alguna de las articulaciones del robot, por lo general a partir de información de esfuerzos . En otros casos el aprendizaje es a alto nivel, como en los procesos de ensamblado y desensamblado automáticos . En cuanto a los métodos de aprendizaje, cabe distinguir fundamentalmente entre aquellos en los que no es necesaria intervención humana o métodos autónomos y aquellos en los que sí lo es o métodos supervisados. En este trabajo se presenta un sistema de aprendizaje supervisado centrado en el control a alto nivel del agarre de objetos.
[Show abstract][Hide abstract] ABSTRACT: In order to be fully autonomous, a mobile robot must possess the skill of finding its location in a particular environment. In other words, while navigating, a mobile robot must be capable of finding its location in a map of the environment (i.e. its pose < x, y, θ >), otherwise the robot will not be able to complete its task. The problem becomes specially challenging if the robot does not possess any external measure of its global position. Naively, the position/orientation of the robot can be determined using odometry sensors or inertial systems. However, these sensors lack of accuracy when used for long periods of time, due to wheel slippage, drifts and other problems. Localization techniques are used instead, in order to find the position of the mobile agent in the space. The great majority of localization methods rely on finding salient characteristics sensed by the robot and relating them with a map of the environment. In this paper we present a localization method based on the Monte Carlo algorithm.
[Show abstract][Hide abstract] ABSTRACT: Evaluation through exams based on multiple choice questions has an important advantage: the results are completely objective. However, this kind of evaluation is seldom used due to its main drawback: cheating, or copying the answers from a classmate, is much easier for the students. We present a simple software tool that manages multiple choice exams. The main idea is automating the generation of different exam versions by shuffling all question versions. First, a Latex document must be created, with the main structure of the exam and the different versions of each question delimited with markers. Then, a simple Matlab script (easily portable to other programming languages) produces a new Latex file where the different versions of each question are combined, resulting in a huge number of exam versions. A separate Latex file is also created, in order to be used by the docent, helping him/her in the correction process. Finally, the appropriate Latex commands are run in order to compile the files and produce two pdf outputs: one for the students (all the version of the exam to be solved) and one for the docent (which helps in the correction process). The Matlab script, as well as usage instructions and examples are freely available from our website (http://lcsi.umh.es).
[Show abstract][Hide abstract] ABSTRACT: An appearance-based visual recognition system for discrimination between free-form objects is presented. The system has been developed in order to be run in a scene where the number of patterns is considerably large and they are not stable, as new patterns are introduced and others are eliminated along the time. In appearance - based systems, the learning step requires the acquisition of large image sets and an intensive computational cost due to the feature extractors. Generally, learning is performed off-line so computing time is not a problem. However, with varying patterns, learning has to be performed several times and under such circumstances the training time is relevant. The proposed method uses random projection for dimensionality reduction and a k-NN classifier to identify the object correctly. We show that despite the computational simplicity of random projection, it allows to keep enough information of the original appearance-based vectors in order to distinguish between the patterns. Experimental results obtained with several image databases show the validity of the approach.
[Show abstract][Hide abstract] ABSTRACT: Nowadays, Internet-based techniques have become a powerful tool in all the fields of engineering education. This paper presents a distributed platform that allows the students to access the robots available in the laboratory through Internet. With this platform, a remote environment is created so that the students can carry out different experiments over several available equipments in the lab, with a flexible schedule. This way, the users can create algorithms of basic reactive control and test them on real robotic platforms. Currently, there are three kinds of robots available. The main objective of the work is create a common platform that allows the access to all of them using a common interface of communication in a transparent way from the point of view of the user.