Stefan Gheorghe Pentiuc

Hand posture recognition for human-robot interaction

In this paper, we describe a fast and accurate method for hand posture recognition in video sequences using multiple video cameras. The technique we propose is based on the head detection, skin detection and human body proportions in order to recognize commands from real-time video sequences. Our technique is also a robust one in order to deal with changes of lighting. The experimental results show that it can be used in various vision-based applications that require real-time detection and recognition of hand posture.

Towards the Optimized Personalized Therapy of Speech Disorders by Data Mining Techniques

Various speech disorders or language impairments can affect the whole life of a person. Discovered and treated in time, they can be corrected, most often in childhood. The use of information technology in order to assist and supervise speech disorder therapy allows specialists to collect a considerable volume of data about the personal or familial anamnesis, regarding various disorders or regarding the process of personalized therapy. These data can be the foundation of data mining processes that show interesting information for the design and adaptation of different therapies in order to obtain the best results in conditions of maximum efficiency. The aim of this paper is to make a short analyze of the use opportunity of the data mining techniques in order to improve the personalized therapy of speech disorders framework. We also present Logo-DM, a data mining system designed to be associated with TERAPERS system in order to provide information based on which one could improve the process of personalized therapy.

Robust 3D Hand Detection for Gestures Recognition

The aim of this paper is to present a fast, robust and adaptive method for real-time 3D hand detection. Unlike traditional 2D approaches which are based on skin or features detection, the proposed method relies on depth information obtained from a stereoscopic video system. As a result, it provides the 3D position of both hands. It also deals with changes of lighting and it is capable of detecting the hands from long distances. The experimental results showed that our method can be successfully integrated in various and complex vision-based systems requiring real-time recognition of hands gestures in 3D environments.

Control of a Mobile Robot by Human Gestures

The control of a mobile robot based on natural human gestures represents a challenge and a necessity. In this paper we will try to present an innovative system that controls a mobile robot using human gestures performed with arms. The human operator acts in the front of a Microsoft Kinect sensor [1] that identifies the skeleton and transmits the coordinates of its joint points to the PC. It follows the gestures recognition process, and transmission to the mobile robot. We succeed to integrate the skeletal tracking, gesture recognition and the control of the mobile robot in an unique application. With this application a number of experiments were deployed, that permit us to evaluate the performances of the proposed interaction system. The overall system evaluation has been made using the following performance parameters: classification accuracy, error rate, precision, recall, sensitivity and specificity.

Multilevel Parallelization of Unsupervised Learning Algorithms in Pattern Recognition on a Roadrunner Architecture

The aim of the paper is to present a solution to the NP hard problem of determining a partition of equivalence classes for a finite set of patterns. The system must learn the classification of the weighted patterns without any information about the number of pattern classes, based on a finite set of patterns in a metric pattern space. Because a metric is not suitable in all the cases to build an equivalence relation, an ultrametric is generated from indexed hierarchies. The contributions presented in this paper consists in the proposal of multilevel parallel algorithms for bottom-up hierarchical clustering, and hence for generating ultra-metrics based on the metrics provided by the user. The algorithms were synthesized and optimized for clusters having the Roadrunner architecture (the first supercomputer that breaks 1PFlops barrier [1]).

Towards a multimodal emotion recognition framework to be integrated in a Computer Based Speech Therapy System

Emotion recognition has become a “must have” for all system that want to inspire users confidence and to interact in a friendly and familiar way. In this paper we propose an improved CBST (Computer Based Speech Therapy System) architecture by using multimodal (i.e. paralanguage, visual, and physiological parameters) emotion recognition techniques. Most research on emotion recognition using speech analysis so far has focused on adult subjects, with a good pronunciation. However, little research has been conducted on adapting classical affect recognition techniques in “narrow areas” such as children speech therapy, where emotions play a key role. So, our paper aims to deal with the assessment of the affective state of the children with speech disorders. A brief literature review is presented, exploring the recent work in the area. New hypothesis are formulated in order to identify the limits of using classical emotion recognition techniques in this special conditions. An original framework to be integrated in the CBST architecture is also outlined. The proposed framework can be seen as an extension of a CBST but will be flexible to other learning systems too.

Using a Fuzzy Emotion Model in Computer Assisted Speech Therapy

Affective computing – machine’s ability to recognize and simulate human affects – has become a main research field for Human Computer Interaction. This paper deal with emotion recognition within a CBST (Computer Based Speech Therapy System) for preschoolers and young schoolchildren. Identifying the emotions of children with speech disorders during the assisted therapy sessions requires an adaptation of classical recognition techniques. That is why, in our article we focus on finding and testing the best emotion representation model to be used in this narrow field. An experiment that validates our proposed approach and indicates the probabilistic coefficient matrix is also presented. The proposed emotion recognition framework can be seen as a future extension of our CBST – Logomon.

Analysis of Huge Volume of Documents on a Hybrid HPC Cluster

The analysis of huge volumes of data has always required high computational ability. Implementing the clustering and classification algorithms upon an ordinary computer will not be efficient, since this shows limitations as regards the computational ability and available memory. Within this paper, an algorithm of clustering the huge volumes of data on a hybrid HPC cluster, is proposed. This cluster has a RoadRunner architecture. It includes 48 QS22 blade servers, each of them having 2 PowerXCell 8i processors. The algorithm proposed within the present paper will carry out the clustering of a number of documents, by using a parallel version of the k-means algorithm, implemented upon the RoadRunner hybrid architecture.