Objective Evaluation Of Driving Fatigue By Using Extraction Of Pupillary Fluctuations

Fatigue driving, drunk driving and violated driving are composed ofthree major reasons for the traffic accident. Drunk driving and violateddriving can be determined by alcohol test and video monitoring, respectively.However, to our knowledge, there have not been a method that can be used toobjective evaluate driving fatigue in traffic administration, thus it isnecessary to develop a system for traffic police to objectively evaluate thedrowsiness of a driver in order to avoid casualties and property lossescaused by fatigue driving.The rationale in this thesis is based on the fact that the fluctuation ofpupil is innervated by autonomic nervous system whose activity is knownto change in parallel with drowsiness. A set of digital video capture systemas well as a series of software were developed to acquire pupil diameter as afunction of time under the case of the spontaneous fluctuations orlight-evoked fluctuations, respectively. Furthermore, we analyze the timedomain and frequency domain characteristics of these fluctuations to obtainsome effective parameters that could be used to identified driving fatigueobjectively. The video datasets obtained by home-made system include14alertcases and13fatigue cases of16subjects. All the datasets of every case areconsist of spontaneous fluctuations and light-evoked fluctuations. The pupildiameters in each frame of video are extracted precisely by using themodified curvature algorithm. And then, the curves of the change of pupildiameters as a function of time were produced to further acquire satisfiedoutcomes by analyzing the time and frequency domain of these curves.The result shows that:(1) Time domain: no matter in spontaneous orevoked cases, the pupillary unrest index has a significant difference; thepercentage of time that eyelid80%closure over in spontaneous cases showsa great reliability; the time of pupil constriction, and duration time under thelight stimulation of1sec have significant differences.(2) Frequency domain:in spontaneous cases, power spectrum in0-0.8Hz frequency band,proportions of frequency spectrum amplitude and power spectrum in0.1-0.2Hz frequency band have significant differences; in evoked cases, frequencyspectrum amplitude and power spectrum in0-0.8Hz frequency band,proportions of frequency spectrum amplitude and power spectrum in0.1-0.2Hz frequency band have statistically significant differences.