Neil Hernández-Gress

A comparison of road departure warning systems on real driving conditions

We present a comparison between different warning strategies for onboard rumble strips in order to prevent road departure events. We compare time-to-lane-crossing (TLC) approach and some implementations of variable rumble strip (VRBS). The framework of this study is the AWAKE European research project and the PREDIT French research program. The value of such a system is demonstrated by some field experiments and its validation by medical and self-diagnoses.

Driver vigilance monitoring, a new approach

In this paper, we introduce an original advanced driver assistance system, in order to tackle the Hypovigilance diagnosis problem. Some constraints of the system are: 1) the supervision and monitoring of the driver-vehicle-environment system is a complex problem due to particularities such as stochastic, personal and dynamic; 2) The solution must be based on a non-intrusive system, and 3) able to diagnose and predict Hypovigilance from peripheral on-board sensors. The proposed diagnosis system is based on intelligent signal processing algorithms and real time methodologies as the wavelets analysis and the support vector machines learning approach for density estimation. The framework of this study is the AWAKE European research project. The value of such a system is demonstrated by some field experiments and its validation by means of EEG/EOG physiological measures.

SVM clustering for identification of takagi-sugeno fuzzy models

Abstract In recent years, the importance of the construction of fuzzy models from measured data has increased. Nevertheless, the complexity of real-life process is characterized by nonlinear and non-stationary dynamics, leaving so much classical identification techniques out of choice. In this paper, we present a new fuzzy clustering algorithm for the parameterization of Takagi-Sugeno (TS) fuzzy models using Support Vector Machines (SVM) for density estimation. We show the properties of the identification procedure on a waste-water treatment database.

IMU, INTEGRATED MONITORING UNIT OF THE SAVE DIAGNOSTIC SYSTEM

Abstract In this paper, the diagnostic part of the SAVE system is described. It is composed of three main diagnostic subsystems 1) behavioural diagnosis, 2) physical diagnosis and 3) critical diagnosis. Each subsystem uses different sensors on-board the vehicle and the information is placed in a hierarchical order and mixed to obtain a general diagnosis. A model (black-box), learned off line, is compared on line to the actual drivers performance. Owing to the inherent difficulty of this problem, the model is created by using Statistical and Artificial Intelligence algorithms (Neural Networks and Fuzzy Logic). Laboratory experiments were made and yielded a success rate in excess of 95%. On-line experiments have started using the SAVE demonstrators in which the system is able to differentiate between the driving behaviour of two persons with a success rate of 98% under the same real traffic conditions.

Using Artificial Neural Networks (ANNs) in Detection Methodologies. First Detection Results Concerning a DMDP Antenna Structure

Abstract The objective of this paper is to present detection results, produced from experiments in a Dual Mode Dual Polarity (DMDP) antenna structure, using a general detection methodology based on Statistics and Artificial Neural Networks (ANNs), in order to detect possible abnormalities within the structure. The antenna is supposed in three states (healthy, minus2bolts, minus1mass). Several experiments, both on healthy and damaged DMDP antenna structure, were carried out in a free configuration space i.e. the structure is suspended with elastic strings, so that it has rigid body suspension frequencies. Frequency Response Functions (FRFs) were recorded and used to detect the abnormalities. A simple identifier for all frequencies is proposed and statistically studied. This identifier is then used by the proposed methodology to compute the diagnosis. The diagnosis performance is 100% for all the treated experiments (in laboratory).

Statistics Methodology for Automatic Detection Vigilance in Real Time: Automotive Application

This paper describes an investigation of different methodological approaches used in driver monitoring and detection copilot systems. Three directions are identified: driver monitoring in terms of mistakes, the evaluation time of the drivers reaction, and a driving procedural analysis.