Vehicle control and drowsiness - VTI

More documents

Recommendations

Info

mated and accurate TLC minima occur. The method of TLC estimation basedsolely on lateral distance and lateral velocity gives poor results on these points.Summarizing, the authors recommend using the approximation method based onlateral distance, lateral velocity, and the second derivation of lateral position (thus,the first and second derivation of lateral position) as an approximation of TLC inreal world driving. A more precise method to calculate TLC is presented in thethesis by Chiu-Geng Lin [54]; methods used are among others Kalman Filtering.It has to be noted that the lateral position data has to be sampled with at least10 Hz in order to be usable for TLC calculations. Time-to-line crossing was usedin the SAVE study [45] as a performance measure. The calculations were madebased on 30 seconds of data.Renner and Mehring [19] also used TLC, in their report it was used to initiatelane departure warning. Extensive work related to lane keeping and lane departurewarning (related to TLC and drowsiness) can be found in the Ph.D. thesis byBatavia [55]. In the report by Pomerleau et al. [56] different algorithms tocalculate TLC are presented and evaluated in their performance (see appendix 2).The extensive report also illustrates the results of the different algorithms relatedto TLC and lane keeping.In order to calculate some form of TLC it is needed to estimate the vehicle’stravel path, this prediction can be made in several ways, as exemplified in thework of Batavia et al. [57] Methods for calculation of TLC are for examplememory based learning and kinematics projection. In memory based learningcurrent lane position and lateral velocity is used to look up the most likely futurelane position based on previous training data. Kinematics prediction works asfollowing: given the current vehicle state, which consists of lane position andlateral velocity, compute what the lateral position of the vehicle will be in tseconds, assuming constant lateral velocity, using lp’=lp+(t*lv), where lp’ ispredicted lane position, lp is current lane position, lv is lateral velocity, and t is theprediction time step. Results of the methods by Batavia et al. in terms ofperformance of the TLC prediction are shown in figure 12 and figure 13. Asconcluding remarks for the results from Batavia, it can be said that a number ofdifferent algorithms and methods have being used to study the problematic ofcalculating TLC. The results of the calculations are somewhat similar, and thefinal scope of the calculation determines which method is used.30 VTI meddelande 922A
Figure 12 Prediction results for 50 seconds of driver_10 using kinematicsprediction. The blue (dark) is actual future lane position; the red (bright) ispredicted.Figure 13 Prediction results for 50 seconds of driver_10 memory based learning.The blue (dark) is actual future position, the red (bright) is predicted future laneposition.In the thesis by De Waard ([47], also cited in relation to standard deviation oflateral position, see chapter 4.4 Lateral position) time-to-line crossing wasmentioned to be related to state of fatigue of the driver. Table 4 shows results forminimum and median TLC for fatigued drivers and a baseline (driver notfatigued).Table 4 Time-to-line crossing times in seconds for fatigued drivers compared tobaseline. Results for state "fatigue" and the different TLC are statistically significant.left handmedian TLCleft handminimum TLCright handmedian TLCright handminimum TLCbaseline 5,31 1,89 3,79 1,58fatigue 3,95 1,71 3,11 1,30VTI meddelande 922A 31
Page 1 and 2: VTI meddelande 922A • 2002Vehicle
Page 3 and 4: Publisher:Publication:VTI Meddeland
Page 5 and 6: ForewordThis study was commissioned
Page 7 and 8: ContentPageSummary 7Sammanfattning
Page 9 and 10: Vehicle Control and Drowsinessby Al
Page 11 and 12: Part one: Literature Study1 Introdu
Page 13 and 14: 1.3 ValidationA relevant problem in
Page 15 and 16: 2 Terms and definitions“Fatigue:
Page 17 and 18: 4 Methods to identify drowsiness in
Page 19 and 20: ight light conditions, when people
Page 21 and 22: Micro-correction in steeringMicro-c
Page 23 and 24: in response to conditions in the ro
Page 25 and 26: driving performance measures. Some
Page 27 and 28: SD lateral positionFigure 10 Standa
Page 29: Different measures have been used i
Page 33 and 34: 1: AVEOBS, which is an observer rat
Page 35 and 36: 5 Commercially available fatigue mo
Page 37 and 38: 5.5 Companies that are planning to
Page 39 and 40: 7 Problems related to automatic dro
Page 41 and 42: 8 Further literature of interestThe
Page 43 and 44: 9.3 Other physiological measuresPhy
Page 45 and 46: Note: the summary only illustrates
Page 47 and 48: Figure 16 Lateral position of a dri
Page 49 and 50: Example 2, subject number 17Figure
Page 51 and 52: 11 Analysis of drowsiness ratingIn
Page 53 and 54: 0,60,5SDLP (meters)0,40,30,2TiredAl
Page 55 and 56: As described in the previous sectio
Page 57 and 58: 109Tiredness876Predicted YTiredAler
Page 59 and 60: 14 Times driver change directionA s
Page 61 and 62: A large number of previous studies
Page 63 and 64: The two groups where compared in or
Page 65 and 66: 16 Frequency AnalysisAn analysis of
Page 67 and 68: The large power variation between i
Page 69 and 70: Fourier transform on the two groups
Page 71 and 72: 16.2 Power Spectrum estimate via Bu
Page 73 and 74: Table 9 Average value of dependent
Page 75 and 76: 18.4 Time to Line Crossing - TLCIn
Page 77 and 78: 19 References1. Wierwille, WW & Wre
Page 79 and 80: 29. PERCLOS: A Valid Psychophysiolo
Page 81 and 82:
54. Chiu-Geng, Lin: Lane Sensing an
Page 83 and 84:
87. Haworth, NL & Vulcan, AP: Testi
Page 85 and 86:
Appendix 1Page 1 (2)TLC formulaeVTI
Page 87 and 88:
TLC calculationAppendix 2Page 1 (1)
Page 89 and 90:
Appendix 4Page 1 (2)Definition of t
show all

Vehicle control and drowsiness - VTI

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?