Pierre Loslever

Using driver's head movements evolution as a drowsiness indicator

Many studies have been conducted to try to put forward measurable indicators that could help in the design of alerting devices used to reliably monitor severe driver drowsiness. However, most of these indicators are either too sensitive to environment modifications (performance indicators), invasive for the driver (EEG, EOG...) or are complex and expensive to measure (Eye blinks, PERCLOS...). This study aims at proposing a set of indicators related to driver performance and physiology that could be efficient for monitoring driver drowsiness and be easy to measure. Thirty two subjects took part to a long duration simulated highway trip (336 km) during which performance and physiological data were recorded The analysis of these data puts forward two sets of indicators : firstly, a set of performance indicators whose evolutions are consistent with those described in literature. They ensure the detection of drowsiness during the experiment. Secondly, a set of physiological indicators : the head movements dispersion. This has revealed itself to be a very good indicator as it evolves in the same way as the performance ones on long lasting periods while being poorly influenced by the road environment, contrary to most performance indicators.

Obtaining information from time data statistical analysis in human component system studies. (I). methods and performances

Abstract This article states the problem of the time data exploration as the succession of the many analysis paths taken prior to obtaining results, results which are initially in a latent state. The initial data – obtained by an experiment or an observation design – are placed within a hyperparallelepiped HP0 where the directions correspond to factors established a priori and each cell contains a multidimensional signal. An analysis path is thus considered to be a progressive information transformation of the time data including up to five stages – data characterizing, scale transformation, data shaping, statistical analysis method application and result presentation. For each stage, a non-exhaustive set of methods is proposed. To assess the performance of each stage, several methods are suggested. More particularly, the evaluation of the first stage is considered either in terms of a cardinality reduction between the input and output hyperparallelepipeds HP0 and HP1 or in terms of distance comparisons between the cells of HP0 and HP1. A discussion of the main statistical behaviors encountered in the literature is included.

Fuzzy Segmentation for the Exploratory Analysis of Multidimensional Signals: Example From a Study on Driver Overtaking Behavior

This paper explains the key role played by windowing in the preliminary analysis of multifactor and multivariate (MFMV) databases. The explanation is based on the general case of a database featuring quantitative or qualitative measurement variables and a hyperparallelepipedic structure in which the directions correspond to the factors. In order to maintain the MFMV aspects of this data structure, the windowing approach as described in this paper does not reduce the information as much as most of the basic non-windowing summarizing procedures using the standard statistical indicators. First, the data in each cell of the hyperparallelepiped are transformed into membership values that can be averaged over factors, such as time or individuals. Then, these membership values may be potentially investigated into with several graphic techniques; for this paper, multiple correspondence analysis (MCA) was chosen. The presentation fall into two parts. First, a didactic example based on a simulated data set describes the approach in comparison with more traditional approaches, and then a real data set, with multidimensional signals recorded for 34 subjects in 15 experimental overtaking situations, is used to demonstrate the power of the “space windowing/MCA” pair on a large real database. Next, the discussion section weighs out the pros and the cons of using space windowing to perform a preliminary analysis of a large MFMV database in studies of human component systems.

From classic statistical characterization to fuzzy windowing based characterization for the exploratory analysis of miscellaneous time variables: example in the field of car driving studies

The problem of data characterization of quantitative and qualitative measurement scales is stated in the context of an exploratory multivariate statistical analysis. An example from a car driving study is considered where the quantitative data correspond to the car and head movements, while the qualitative data correspond to objects being viewed--road, bridge, sign-post, etc. For each of these two sets, the literature is analyzed first in terms of data characterizing methods and relationship obtaining methods. Then we propose to evaluate and compare nine quantitative data characteriing methods: five corresponding to classic statistical indicators, two to crisp space windowing with either two or three windows, and two on fuzzy windowing with either two or three windows. Logically the last method appears as the best (according to our evaluation procedure). Then we propose a bidimensional fuzzy windowing instead of a crisp one to characterize the gaze positions. Finally the multiple correspondence analysis is used to investigate the membership value averages obtained from the characterization stage.

Towards a methodology for selecting good scaling factors for a fuzzy controller

Abstract The work presented concerns an approach for selecting the scaling factors of a single fuzzy controller. This approach has been done using as basic systems the ones presented in an advanced showcase of adaptative controller designs (Masten and Cohen, 1990). The paper presents results through an optimisation technique done on these systems by data analysis method.

A scale fuzzy windowing comparison applied to multivariate descriptive analysis

Observational and experimental data are often investigated into so that the factor effects and/or variables connections can be assessed quickly and easily via inference tests. This article suggests starting the statistical analysis using a 5-step descriptive procedure: 1 Data characterization, 2 Data coding, 3 Data table drafting, 4 Data table analysis and 5 Result presentation. In order to illustrate this preliminary statistical analysis, two data set examples are considered --one from a small simulated system and one from a large mechatronic system--using two different methods: Principal Component Analysis with usual statistical summaries and Multiple Correspondence Analysis with indicators obtained through fuzzy space windowing. In an Intelligent Data Analysis context, the discussion weighs out the pros and the cons of these approaches, prior to using procedures 5-step inference procedures.

Using space windowing for a preliminary analysis of complex time data in human component system studies. Examples with eye-tracking in advertising and car/head movements in driving

Empirical studies of human systems often involve recording multidimensional signals because the system components may require physical measurements (e.g., temperature, pressure, body movements and/or movements in the environment) and physiological measurements (e.g., electromyography or electrocardiography). Analysis of such data becomes complex if both the multifactor aspect and the multivariate aspect are retained. Three examples are used to illustrate the role of fuzzy space windowing and the large number of data analysis paths. The first example is a classic simulated data set found in the literature, which we use to compare several data analysis paths generated with principal component analysis and multiple correspondence analysis with crisp and fuzzy windowing. The second example involves eye-tracking data based on advertising, with a focus on the case of one category variable, but with the possibility of several space windowing models and time entities. The third example concerns car and head movement data from a driving vigilance study, with a focus on the case involving several quantitative variables. The notions of analysis path multiplicity and information are discussed both from a general perspective and in terms of our two real examples.

Spécification d'IHM dans les systèmes critiques: retour d'expérience sur une pratique en enseignement de l'IHM

HCI teaching about critical systems requires the use of pedagogical approaches adapted to the specificities linked to the application field. This paper aims at providing a feedback after nine years of practice in a 4 years course program in Electrical Engineering and Industrial Informatics, particularly in HCI Specification. The pedagogical approach is based on a role playing.

Obtaining information from time data statistical analysis in human component system studies. (II). example with simulated data

Abstract A didactic example based on a set of 10 signals is given to illustrate the diversity in time data exploration. We argue that to assess the performance of the analysis path, the researcher must focus mainly on the information loss in the characterizing stage since it is the starting point of the path, thus conditioning its potential results. Several statistical analysis paths with different information loss levels are suggested, and three paths are investigated. The first two are exploratory and based on a space-time fuzzy windowing and Multiple Correspondence Analysis. The third path is confirmatory, based on a non-parametric test to see if the previous results can be generalized. A discussion dealing with the objectivity and subjectivity aspects inherent to a statistical analysis is then proposed.

Classification of evaluation strategies of the term “large” and the connective “and” through correspondence factor analysis

Abstract The problems caused by the evaluation of the concepts of height and of the conjunction ‘and’ by human operators is considered in this paper. Therefore, an experiment based on graphical views, such as bargraphs, pairs of bargraphs and rectangles, and linguistic assertions, is proposed. Although these problems have sometimes been studied through fuzzy sets theory to find some models, a new way of treatment for the data obtained is used: correspondence factor analysis. This method allows one to find classes of subjects with differents behaviors brought out by factorial planes.