Catálogo de publicaciones - libros

In this article I focus on a simple education game Penguin Quart which is designed to use a speech dialogue. A genesis of it was motivated by effort to try a digit speech recognition in a real environment. The game was developed universally not only to educate but also to collect digit speech samples and to improve its recognition accuracy.

Face recognition systems detect faces in moving or still images and then recognize them. However, face detection is not an error-free process, especially when designed for real-time systems. Thus the face recognition algorithms have to operate on faces that are not ideally framed. In this paper we analyze quantitatively the impact of face detection errors on six different face recognition algorithms. Hence, we propose a matching of face recognition algorithms with face detector performance, which can be used for a system based on the expected performance of the face detector.

This paper addresses the problem of segmenting small group meetings in order to detect different group configurations and activities in an intelligent environment. Our approach takes speech activity detection of individuals attending a meeting as input. The goal is to separate distinct distributions of speech activity observation corresponding to distinct group configurations and activities. We propose an unsupervised method based on the calculation of the Jeffrey divergence between histograms of speech activity observations. These histograms are generated from adjacent windows of variable size slid from the beginning to the end of a meeting recording. The peaks of the resulting Jeffrey divergence curves are detected using successive robust mean estimation. After a merging and filtering process, the retained peaks are used to select the best model, i.e. the best speech activity distribution allocation for a given meeting recording. These distinct distributions can be interpreted as distinct segments of group configuration and activity. To evaluate, we recorded 6 small group meetings. We measured the correspondence between detected segments and labeled group configurations and activities. The obtained results are promising, in particular as our method is completely unsupervised.

In this paper, we present a real-time surveillance system that is suitable for the indoor environment. The system is designed to detect, track and recognize the behavior of humans, using a single static camera. Background subtraction is applied to extract moving objects; these objects are tracked using linear approximation. Shadow regions are detected and removed using linear dependence and spatial connectivity properties of the shadow regions. Pattern matching and TDL (Two Dimensional Logarithmic) search approach are used to solve the problem of the occlusion of objects and depth reasoning. Behaviors of moving objects are detected by examining the sequence of shapes extracted from the scene. Shapes of moving objects are interpreted as characters of an alphabet. Each character represents a class of similar blob shapes classified using K-Means clustering. The model is used to recognize behaviors in an office with promising results.

Most pedestrian detection systems are built based on computer vision technology and usually are composed of two basic modules: object detection module, and recognition module. This paper presents an efficient filtering module, which works between the two basic modules, based on extracting the 3-dimensional information from single frame images. The filter module removes the noisy objects extracted by object detection module and thus reduces the burden of the recognition module. 3-D information, such as height, width and distance are extracted from single frame images. Using this information, a Bayesian classifier is employed to implement the filter. The main contribution of this filter module is that it removed about 30% noisy objects detected by the object detection module. The total computing cost and error detection rate is reduced when this filter module is used in the pedestrian detection system.

The proliferation of mobile computing devices and local-area wireless networks has fostered a growing interest in location-aware systems and services. Additionally, the ability to let a mobile device determine its location in an indoor environment at a fine-grained level supports the creation of a new range of mobile control system applications. Main area of interest is in model of radio-frequency (RF) based system enhancement for locating and tracking users of control system inside buildings. The locating and tracking of users is useful for rescue people to find closer way in unknown building. The software can navigate these people through the unknown space to accident place. The experimental framework prototype uses a WiFi network infrastructure to let a mobile device determine its indoor position as well as to deliver IP connectivity. Experiments show that location determination can be realized with a room level granularity.

We present a hybrid system based on a combination of Neural Networks and rule-based matching systems that is capable of detecting network-initiated intrusion attacks on web servers. The system has a strong learning component allowing it to recognize even novel attacks (i.e. attacks it has never seen before) and categorize them as such. The performance of the Neural Network in detecting attacks is very good with success rates of more than 78% in recognizing new attacks. However, because of an alarmingly high false alarm rate that measures more than 90% on normal HTTP traffic carrying image uploads we had to combine the original ANN with a rule-based component that monitors the server’s system calls for detecting unusual activity. A final component combines the two systems to make the final decision on whether to raise an intrusion alarm or not. We report on the results we got from our approach and future directions for this research.

This article deals with the choice of individual trust models adapted to networks. We consider trust as a social and effective multi-agent process. We introduce the notion of trust networks viewed as a set of one-to-one trust relationships, we wonder which trust model should be chosen to build and exploit it. We extract five criteria for comparison of trust models. We then evaluate two trust models and discuss what could be a relevant trust model in a multi-agent setting.

XML technologies have been recently extensively used in IP based network management, where they have been proven capable of alleviating the SNMP shortcomings in configuration management. Our XMLNET system described in [] has demonstrated that XML systems can greatly facilitate the development of network management applications even in complex heterogeneous multi-vendor networks. In this paper, we present extensions to the XMLNET architecture, with a view to managing not only network devices, but also middleware and hardware resources used in the scope of ubiquitous computing. Ubiquitous computing services are supported by a highly distributed and heterogeneous infrastructures comprising a wide range of sensors (e.g., cameras, microphones, motion sensors, temperature sensor), a well as middleware components (e.g., recognition algorithms, perceptual components). The introduced extensions to XMLNET for ubiquitous computing environments, aim at lever-aging the merits of XMLNET for the inherently complex configuration management operations entailed in pervasive and ubiquitous computing applications.

This paper presents an interactive constraint based system that simultaneously assist design and evaluation. This work is driven by an industrial case dealing with heat treatment operation. The first part presents the problem and provide ideas of the solution. Then the knowledge model mixing discrete and numerical constraints is presented. The third section provides filtering elements in order to permit interactive assistance. The last one discusses the designed system. The originality of the proposition lies in the gathering of classical discrete constraints filtering techniques with numerical constraint 2B consistence filtering mechanisms that were necessary to respond to the industrial need.