Back to search

University of Ioannina

Παρακολούθηση ψυχοσωματικής κατάστασης οδηγού με χρήση βιοσημάτων

Abstract

dc:description

Among the most important factors in accident provocation is the human factor. Driver's loss of attention, aggressiveness and inefficient decision making, often lead to hazardous situations. A common reason of all the above is the physiological state of the driver. The last few decades several studies have designated the role of physiological state in driving performance and accident provocation. However, the design of in-vehicle systems, able to assess physiological state with high credibility and in a rather unobtrusive way is still an ambitious goal. This thesis makes a contribution towards the goal of driver's status recognition in a real-car application. Based mainly on the monitoring of driver's physiological signals and using additional sources such as video from driver's face as well as driving environment information, we focused on the detection of driver's fatigue and stress levels. The data used were collected during a long-lasting experimental protocol on a real driving environment and under different weather, traffic and road conditions. During the real-world experiments a set of physiological signals have been recorded, in particular: electrocardiograph (ECG), respiration and electrodermal activity (EDA), from which a number of features from both time and frequency space were extracted. The set of features were examined in terms of their contribution to the detection of fatigue and stress and the most powerful indicators for each state of interest were selected following a feature selection technique. Four different classifiers were employed and the classification results were evaluated in terms of sensitivity, specificity and accuracy. Our results indicate good performance for both fatigue and stress classification problems. The evaluation of features' contribution to the detection of the two states revealed that all information sources contribute equally to the two-class stress classification problem, while a three-class fatigue classification is feasible by using physiological features only, as those are the features with the major contribution. Although the majority of relevant works concentrate on the detection of high levels of fatigue (drowsiness) with our methodology we can achieve also detection of early fatigue stages (low fatigue). To quantify the impact of these early fatigue stages on driving performance, we built a driving simulation environment, which allowed us to monitor driving performance measures (i.e. reaction times) and correlate them with the states under investigation. The outcomes of the simulation study verified the association of the different fatigue levels and the physiological features used for their classification with the degree of impairment in driver's performance. This finding demonstrates the value and potential of an in-vehicle system able to recognize progressive fatigue levels and predict driver's reaction time based on physiological signals monitoring. Unlike fatigue which is a progressive state, driver stress is a more time-variant condition which is highly dependent on the driving context and it is temporarily influenced by events occurring en-route. We call the driving environment-related events (such as overtake, hard braking, junction, curve, etc.) "driving events" and the temporal stress that may impose to the driver, "stress events". The detection of stress events can contribute to the driver's state estimation in the sense that the frequent occurrence of stress events could have more prolonged effects on driver's state, i.e. increase the overall driver's stress level. In our thesis, we developed a methodology for online detection of stress events, using the same set of physiological signals (ECG, Respiration, EDA) as well as driving behavior parameters (position, velocity, acceleration, deceleration) acquired from car equipment (GPS and CAN bus). We achieved 95% accuracy in the detection of stress events which is expected to increase with the introduction of visual information from the road environment. A very important requirement for a system for driver physiological state assessment, is the adaptation of system's parameters on the physiological features of new drivers. We consider a system based on a Gaussian mixture model, and the basic assumption is that the feature distribution of the new driver is a geometric transformation of the feature distribution used for the training of the system. The method is based on the Expectation-Maximization (EM) approach for the estimation of the geometric transformation applied on the initial Gaussian Mixture model. We evaluated the proposed method using artificial dataset. Since EM is a local optimization algorithm, the method does not guarantee the correct solution identification. To confront this limitation, we proposed a multiple start EM algorithm which increased the method performance. Next we extended the model allowing each mixture component to have an individual (local) transformation, and applied a MAP-EM approach for estimating both the common (global) transformation as well as the local ones. We evaluated the proposed method using both artificial and real driver's data, collected in a simulation environment. Apart from adaptation to new drivers, the proposed method can be applied also to other problems, such as image registration and tracking. Based on the fact that driver's state is a significant factor for road safety, we made a step further and designed the building blocks of an advanced driver support systems that exploits driver's state estimation as well as vehicle communication abilities for increasing safety in every day driving. As a first step we developed a simulation environment of vehicle to vehicle (V2V) and vehicle to infrastructure (V2I) communication. Experiments on the simulation environment confirmed that the exploitation of those types of communication can increase driving safety. Moreover, it helped to define minimum requirements for communication characteristics (range, latencies and broadcasting frequency) in realistic scenarios of co-operative driving. The second step was the implementation of two basic subsystems: a) the information handler and b) the decision maker. The information handler applies information fusion of low confidence sensors (other vehicles) and employs Bayesian Networks for the extraction of high-level information, useful for the driver. The decision maker is based upon sampling of Dynamic Bayesian networks for driving risk evaluation and decides upon the best warning strategy, taking into account both environmental conditions and driver's state. Concluding, this thesis studies the assessment of driver's psycho-physiological state exploiting mainly physiological signals as well as other information sources towards the development of a real-application driver support system that contributes to the increase of driving safety.

Degree

thesis:*
Grantor dc:publisher
University of Ioannina
Year dc:date
2009

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Rigas, Georgios
  • Ρήγας, Γεώργιος

Subjects

dc:subject × 12

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/18541

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Rigas, Georgios; Ρήγας, Γεώργιος. Παρακολούθηση ψυχοσωματικής κατάστασης οδηγού με χρήση βιοσημάτων. University of Ioannina, 2009. http://hdl.handle.net/10442/hedi/18541