Professional Documents
Culture Documents
2, JUNE 2012
Abstract—The ability to classify driver behavior lays the foun- and/or observe the danger involved in such situations [3]. These
dation for more advanced driver assistance systems. In particular, data suggest that driver assistance or warning systems may
improving safety at intersections has been identified as a high have an appropriate role in reducing the number of accidents,
priority due to the large number of intersection-related fatalities.
This paper focuses on developing algorithms for estimating driver improving the safety and efficiency of human-driven ground
behavior at road intersections and validating them on real traffic transportation systems. Such systems typically augment the
data. It introduces two classes of algorithms that can classify driver’s situational awareness and can also act as collision
drivers as compliant or violating. They are based on 1) support mitigation systems [4].
vector machines and 2) hidden Markov models, which are two very Research on intersection decision support systems has be-
popular machine learning approaches that have been used suc-
cessfully for classification in multiple disciplines. However, existing come quite active in both academia and the automotive industry.
work has not explored the benefits of applying these techniques to In the US, the federal DOT, in conjunction with the California,
the problem of driver behavior classification at intersections. The Minnesota, and Virginia DOTs, as well as several U.S. research
developed algorithms are successfully validated using naturalistic universities, is sponsoring the Intersection Decision Support
intersection data collected in Christiansburg, VA, through the U.S. project [3], [5] and, more recently, the Cooperative Intersection
Department of Transportation Cooperative Intersection Collision
Avoidance System for Violations initiative. Their performances are Collision Avoidance Systems (CICAS) project [6]. In Europe,
also compared with those of three traditional methods, and the the InterSafe project was created by the European Commission
results show significant improvements with the new algorithms. to increase safety at intersections. The partners in the InterSafe
Index Terms—Driver behavior, driver warning systems, inten- project include European vehicle manufacturers and research
tion prediction. institutes [7]. Both projects try to explore the requirements,
tradeoffs, and technologies required to create an intersection
I. I NTRODUCTION collision avoidance system and demonstrate its applicability on
TABLE I
CLASSIFICATION CATEGORIES
condition happens first. The time and distance thresholds of temporal property is what humans use to decide whether a
are chosen such that the host driver has enough time to vehicle will be compliant [30]. The vehicle’s T T I is defined
react to the warning. A detailed analysis of the choice of simply as
T T Imin and dmin is presented in Section V-B.
3) The target vehicles are tracked as early as possible, but r
TTI = (1)
their classification as violating or compliant is based on v
measurements taken in the Tw time window (Fig. 2).
where v is the vehicle’s current speed, and r is its distance to the
Different values of Tw are analyzed in the developed
stop line. For this classifier, the T T I value is computed when
algorithms; a larger Tw brings a longer measurement
the vehicle’s deceleration crosses some predefined threshold,
“memory” at the expense of an additional computation
indicating the onset of braking. Then, the driver is classified as
requirement. A large Tw might also include irrelevant
a violator if T T I < T T Ireq , where T T Ireq is the time given
measurements when the vehicle is very far from the
for a driver to stop safely after the onset of braking [24]. This
intersection. Finally, note that a target vehicle that stops in
static parameter can be adjusted to change how conservative the
or before the Tw window is directly labeled as compliant.
algorithm is in its classifications.
2) Static RDP: Other work has used a classifier based on
III. A LGORITHMS RDP [15], [16]. RDP gives the deceleration (in g) needed for
the vehicle to stop safely given its current distance and speed.
Classifying human drivers is a very complex task because of
It is defined as
the various nuances and peculiarities of human behaviors [25].
Researchers have shown that the state of a vehicle driver lies in
v2
some high-dimensional feature space [26]. RDP = (2)
Basic classification is traditionally performed by identifying 2rg
simple relationships or trends in data that define each class.
This includes using techniques such as model fitting and re- where r and v are as previously defined, and g is the grav-
gression to identify classification criteria [27]. However, by itational acceleration constant. For a given RDP threshold
only considering simple relationships, these approaches are RDPwarn , a warning distance is computed as
limited in their ability to accurately classify complex data
v2
where the classes may be defined by a variety of factors. To rwarn = . (3)
overcome this limitation, two approaches to classification have 2RDPwarn
been developed in the machine learning community.
Discriminative approaches, such as SVMs, are typically used The vehicle is then classified as a violator if at any time
in binary classification problems, which make them appropriate r < rwarn , i.e., if its required deceleration is greater than the
for the classification of compliant versus violating drivers. selected RDP threshold.
SVMs have several useful theoretical and practical characteris- 3) SDR: A more complex classification strategy based on
tics [28]. We highlight two of them: 1) training SVMs involves fitting regression curves is described in [16] and [31]. This
an optimization problem of a convex function, thus the optimal approach takes a set of speed and distance measurements from
solution is a global solution (i.e., no local optima); 2) the vehicles that are known to be compliant, discretizes the speeds,
upper bound on the generalization error does not depend on the and collects the distance measurements at each speed into a set
dimensionality of the problem. of bins based on percentiles. A regression curve of the form
Classification is often also performed using generative ap-
proaches, such as HMMs, to model the underlying patterns in rwarn = av b + c (4)
a set of observations and explicitly compute the probability
of observing a set of outputs for a given model [29]. HMMs is then fit for each bin. These SDR curves attempt to identify
are well suited to the classification of dynamic systems [9], relationships between speed and distance that can discriminate
such as a vehicle approaching an intersection. The states of the between compliant and violating drivers. For a given curve
HMM define different behavioral modes based on observations, (corresponding to a certain percentile of compliant driver tra-
and the transitions between these states capture the temporal jectories), the vehicle is classified as a violator if at any time
relationship between observations. r < rwarn , i.e., if it is closer to the stop line than expected for
a compliant vehicle at its current speed. Selecting a curve cor-
responding to higher percentile bins yields a more conservative
A. Traditional Methods
classifier.
This section presents three notable techniques from the lit- The version of this algorithm in [16] includes two additional
erature for classifying drivers based on simple relationships layers that declare a vehicle to be compliant if its deceleration
between observations: static T T I [24], static RDP [15], and is below some fixed threshold (e.g., due to braking) or if its
SDR [16]. These algorithms are widely used and provide a velocity is below some fixed threshold (e.g., to permit rolling
baseline for the performance analysis in Section VI. stops). However, only the deceleration layer is considered here
1) Static TTI: One of the most intuitive approaches to clas- as rolling stops are typically not associated with signalized
sification is to use the vehicle’s T T I. It is thought that this type intersections.
AOUDE et al.: DRIVER BEHAVIOR CLASSIFICATION AT INTERSECTION AND VALIDATION ON LARGE DATA SET 727
βi (K) = 1 (22)
n
βj (k) = Tij ej (xk+1 )βi (k + 1). (23)
j=1
Fig. 5. Basic HMM λ(T , t, e) with n = 3 states, transition probabilities Tij , Using the terms αi (k) from the forward algorithm and βi (k)
and emissions ei .
from the backward algorithm, the probability of being in state
si at time k given the observations x is given by
distributed around some typical values (specified by the means
and variances). αi (k)βi (k)
A common task with HMMs is determining how well a γi (k) = P (qk = si |x, λ) = n . (24)
i=1 αi (k)βi (k)
given model λ(T , t, e) fits a sequence of observations x =
x1 , . . . , xK . This can be quantified as the probability of observ- Then, the probability of being in state si at time k and state sj
ing x given λ, P (x|λ). The forward algorithm is an efficient at time k + 1 is given by
method for computing this probability [29] and is defined as
follows. Let αi (k) be given by ξij (k) = P (qk = si , qk+1 = sj |x, λ)
αi (k)Tij ej (xk+1 )βj (k + 1)
αi (k) = P (x1 , . . . , xk , qk = si |λ) (16) = n n . (25)
i=1 j=1 αi (k)Tij ej (xk+1 )βj (k + 1)
Fig. 7. Regression curves used as decision thresholds for the SDR algorithm.
Fig. 10. ROC curves for all five algorithms with insets showing area of interest around 5% false positives. (a) ROC curves for 10 000/1000 basic test at
T T Imin = 1.0 s. (b) ROC curves for 5000/1000 m-fold test at T T Imin = 1.0 s. (c) ROC curves for 10 000/1000 basic test at T T Imin = 1.6 s. ROC curves
for 5000/1000 m-fold test at T T Imin = 1.6 s. (d) ROC curves for 10 000/1000 basic test at T T Imin = 2.0 s. (e) ROC curves for 5000/1000 m-fold test at
T T Imin = 2.0 s.
and HMM-based classifiers, would also see a substantial per- work [16]. This ensures that the new algorithms developed in
formance improvement. However, since the objective is to this work are evaluated accurately against the baseline per-
provide an early warning, T T Imin must be selected to be formance set by these traditional methods. The basic general-
larger than the typical driver response time, as described in ization and m-fold cross-validation tests show the consistent
Section V-B. improvement the SMV-BF and HMM-based algorithms pro-
The general trends observed with these traditional methods vide in classifying a new set of trajectories. This validates
are also consistent with the behaviors observed in previous the two approaches and demonstrates the flexibility embedded
AOUDE et al.: DRIVER BEHAVIOR CLASSIFICATION AT INTERSECTION AND VALIDATION ON LARGE DATA SET 735
TABLE IV R EFERENCES
TRUE POSITIVE RATES FOR EACH ALGORITHM
FOR 5% F ALSE P OSITIVES [1] W. D. Jones, “Keeping cars from crashing,” IEEE Spectrum, vol. 38, no. 9,
pp. 40–45, Sep. 2001. [Online]: Available: http://dx.doi.org/10.1109/
6.946636.
[2] Fatality analysis reporting system encyclopedia, National Highway Traffic
Safety Administration, Washington, DC. [Online]. Available: http://www-
fars.nhtsa.dot.gov/Crashes/CrashesLocation.aspx
[3] B. Bougler, D. Cody, and C. Nowakowski, “California intersection deci-
sion support: A driver-centered approach to left-turn collision avoidance
system design,” Univ. Calif. Berkeley, Berkeley, CA, Tech. Rep., 2008.
[4] G. S. Aoude, B. D. Luders, D. S. Levine, K. K. H. Lee, and J. P. How,
“Threat assessment design for driver assistance system at intersections,”
in Proc. IEEE Conf. Intell. Transp. Syst., Madeira, Portugal, Sep. 2010,
pp. 1855–1862.
[5] C. Y. Chan and B. Bougler, “Evaluation of cooperative roadside and
vehicle-based data collection for assessing intersection conflicts,” in Proc.
IEEE Intell. Veh. Symp., 2005, pp. 165–170.
[6] M. Maile, F. A. Zaid, L. Caminiti, J. Lundberg, and P. Mudalige, “Cooper-
ative intersection collision avoidance system limited to stop sign and traf-
fic signal violations,” Nat. Highway Traffic Safety Admin., Washington,
DC, Midterm Phase 1 Rep., 2008.
[7] K. Fuerstenberg, “New European approach for intersection safety—The
EC-project INTERSAFE,” in Proc. IEEE Intell. Transp. Syst., 2005,
pp. 432–436.
[8] D. D. Salvucci, “Inferring driver intent: A case study in lane-change
detection,” in Proc. 48th Annu. Meeting Human Factors Ergonom. Soc.,
2004, pp. 2228–2231.
[9] N. Oliver and A. P. Pentland, “Graphical models for driver behavior recog-
nition in a smartcar,” in Proc. IEEE Intell. Veh. Symp., 2000, pp. 7–12.
[10] T. Gates, D. Noyce, L. Laracuente, and E. Nordheim, “Analysis of driver
in their design. Furthermore, the ROC analysis and test- behavior in dilemma zones at signalized intersections,” Transp. Res. Rec.
ing of multiple T T Imin values gives the designer or engi- J. Transp. Res. Board, vol. 2030, no. 1, pp. 29–39, 2007.
[11] H. Rakha, I. El-Shawarby, and J. R. Sett, “Characterizing driver behav-
neer the flexibility to select the parameters that best match ior on signalized intersection approaches at the onset of a yellow-phase
the requirements for any real-world implementation of these trigger,” IEEE Trans. Intell. Transp. Syst., vol. 8, no. 4, pp. 630–640,
algorithms. Dec. 2007.
[12] L. Zhang, K. Zhou, W. Zhang, and J. A. Misener, “Prediction of red light
running based on statistics of discrete point sensors,” Transp. Res. Rec. J.
Transp. Res. Board, vol. 2128, pp. 132–142, 2009.
VII. C ONCLUSION [13] J. Bonneson and H. Son, “Prediction of expected red-light-running fre-
quency at urban intersections,” Transp. Res. Rec. J. Transp. Res. Board,
This paper has introduced two new approaches for classify- vol. 1830, pp. 38–47, 2003.
ing driver behaviors at road intersections. The first approach, [14] N. Elmitiny, X. Yan, E. Radwan, C. Russo, and D. Nashar, “Classification
analysis of driver’s stop/go decision and red-light running violation,”
which is denoted as SVM-BF, combines an SVM classifier Accident Anal. Prev., vol. 42, no. 1, pp. 101–111, Jan. 2010.
with a BF to discriminate between compliant drivers and [15] V. Neale, M. Perez, Z. Doerzaph, S. Lee, S. Stone, and T. Dingus, In-
violators based on vehicle speed, acceleration, and distance tersection Decision Support: Evaluation of a Violation Warning System
to Mitigate Straight Crossing Path Crashes (report no. vtrc 06-cr10).
to intersection. The second approach, which is an HMM-
Charlottesville, VA: Virginia Trans. Res. Council, 2006.
based classifier, uses the EM algorithm to develop two distinct [16] Z. Doerzaph, V. Neale, and R. Kiefer, “Cooperative intersection collision
HMMs for compliant and violating behaviors. To optimize avoidance for violations: Threat assessment algorithm development and
safety while respecting driving acceptance levels, the algo- evaluation method,” presented at the Transportation Research Board 89th
Annu. Meeting, Washington, DC, 2010, Paper 10-2748.
rithms were designed to maximize true positive rates while [17] Z. Doerzaph, “Development of a threat assessment algorithm for intersec-
keeping the false alarm rates below the 5% threshold. The two tion collision avoidance systems,” Ph.D. dissertation, Virginia Polytechnic
algorithms were successfully validated on more than 10 000 Inst. State Univ., Blacksburg, VA, 2007.
[18] D. McNicol, A Primer of Signal Detection Theory. Hillsdale, NJ:
intersection approaches collected in Christiansburg, VA, as part Lawrence Erlbaum, 2004.
of the U.S. DOT CICAS-V initiative. The performances of the [19] Car-2-Car Communication Consortium. [Online]. Available: http://www.
algorithms were also compared with three popular traditional car-to-car.org
[20] Vehicle infrastructure integration consortium (VIIC). [Online]. Available:
approaches consisting of TTI-based, RDP-based, and SDR- http://www.vehicle-infrastructure.org
based algorithms. The results of several generalization tests [21] J. Leonard, J. P. How, S. Teller, M. Berger, S. Campbell, G. Fiore,
showed consistent and significant improvements with the de- L. Fletcher, E. Frazzoli, A. Huang, S. Karaman, O. Koch, Y. Kuwata,
D. Moore, E. Olson, S. Peters, J. Teo, R. Truax, M. Walter, D. Barrett,
veloped algorithms, ranging from a minimum of 10% increase A. Epstein, K. Maheloni, K. Moyer, T. Jones, R. Buckley, M. Antone,
in true positive rates to more than 20% increase when issuing a R. Galejs, S. Krishnamurthy, and J. Williams, “A perception-driven au-
warning 1 and 2 s in advance, respectively. tonomous urban vehicle,” J. Field Robot., vol. 25, no. 10, pp. 727–774,
Oct. 2008.
[22] L. Malta, C. Miyajima, and K. Takeda, “A study of driver behavior under
potential threats in vehicle traffic,” IEEE Trans. Intell. Transp. Syst.,
ACKNOWLEDGMENT vol. 10, no. 2, pp. 201–210, Jun. 2009.
[23] S. Thrun, What We’re Driving at Google’s Official Press Release An-
The authors would like to thank Dr. T. Pilutti, Dr. W. Najm, nouncing the Success of Their Driverless Cars, Oct. 2010. [Online]. Avail-
Prof. J. Leonard, and Dr. L. Fletcher for their valuable input. able: http://googleblog.blogspot.com/2010/10/what-were-driving-at.html
736 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 13, NO. 2, JUNE 2012
[24] R. J. Kiefer, J. Salinger, and J. J. Ference, “Status of NHTSA’s rear-end Georges S. Aoude (M’11) received the B.Eng. de-
crash prevention research program,” presented at the 19th Int. Tech. Conf. gree in computer engineering from McGill Univer-
Enhanced Safety Vehicles, Washington, DC, 2005, Paper 05-0282. sity, Montreal, QC, Canada, in 2005 and the S.M. and
[25] H. M. Mandalia and D. D. Salvucci, “Using support vector machines for Ph.D. degrees in aeronautics and astronautics from
lane-change detection,” in Proc. Human Factors Ergonom. Soc. Annu. the Massachusetts Institute of Technology (MIT),
Meeting, 2005, vol. 49, pp. 1965–1969. Cambridge, in 2007 and 2011, respectively.
[26] D. Wipf and B. Rao, “Driver intent inference annual report,” Univ. From 2005 to 2008, he was part of the SPHERES
California, San Diego, CA, Tech. Rep., 2003. team at MIT, where he designed spacecraft recon-
[27] R. O. Duda, P. E. Hart, and D. G. Stork, Pattern Classification. figuration maneuvers that were performed onboard
New York: Wiley-Interscience, 2001. the ISS. He is with the Department of Aeronautics
[28] V. Vapnik, The Nature of Statistical Learning Theory. New York: and Astronautics, MIT. He received the NSERC and
Springer-Verlag, 1995. FQRNT Ph.D. (2007–2010) and Masters (2005–2007) scholarships. He also
[29] L. Rabiner, “A tutorial on hidden Markov models and selected applica- received the 2003 Kenneth Young Lochhead and the 2002 CMC Electronics
tions in speech recognition,” Proc. IEEE, vol. 77, no. 2, pp. 257–286, scholarships. His research interests include intention prediction, threat assess-
Feb. 1989. ment, and real-time path planning under uncertainty.
[30] A. Van der Horst, “A time-based analysis of road user behavior in normal Dr. Aoude is a member of the American Institute of Aeronautics and
and critical encounters,” Ph.D. dissertation, Delft Univ. Technol., Delft, Astronautics.
The Netherlands, 1990.
[31] H. Berndt, S. Wender, and K. Dietmayer, “Driver braking behavior during
intersection approaches and implications for warning strategies for driver
assistant systems,” in Proc. Intell. Veh. Symp., Istanbul, Turkey, 2007, Vishnu R. Desaraju (M’11) received the B.S.E.
pp. 245–251. degree in electrical engineering with Engineering
[32] G. S. Aoude and J. P. How, Using support vector machines and Bayesian Global Leadership honors from the University of
filtering for classifying agent intentions at road intersections, MIT, Michigan, Ann Arbor, in 2008 and the S.M. degree in
Cambridge, MA, Tech. Rep. ACL09-02. [Online]. Available: http://hdl. aeronautics and astronautics from the Massachusetts
handle.net/1721.1/46720 Institute of Technology (MIT), Cambridge, in 2010.
[33] G. S. Aoude, V. R. Desaraju, L. H. Stephens, and J. P. How, “Behavior From 2009 to 2010, he worked on the Agile
Robotics for Logistics project with MIT to develop
classification algorithms at intersections and validation using natural-
semi-autonomous field robotics for the U.S. Army.
istic data,” in Proc. IEEE Intell. Veh. Symp., Baden-Baden, Germany,
He is with the Department of Aeronautics and Astro-
Jun. 2011, pp. 601–606.
nautics, MIT. His current research interests include
[34] SVM Application List, 2006. [Online]. Available: http://www.clopinet.
real-time motion planning, multiagent control, adaptive control, and collision
com/isabelle/Projects/SVM/applist.html
avoidance strategies.
[35] Y. Liang, M. L. Reyes, and J. D. Lee, “Real-time detection of driver
Mr. Desaraju is a member of Eta Kappa Nu.
cognitive distraction using support vector machines,” IEEE Trans. Intell.
Transp. Syst., vol. 8, no. 2, pp. 340–350, Jun. 2007.
[36] T. Ersal, H. Fuller, O. Tsimhoni, J. Stein, and H. Fathy, “Model-based
analysis and classification of driver distraction under secondary tasks,”
Lauren H. Stephens (S’11) is currently an under-
IEEE Trans. Intell. Transp. Syst., vol. 11, no. 3, pp. 692–701, Sep. 2010. graduate student majoring in computer science and
[37] Y. Zhang, W. Lin, and Y.-K. Chin, “A pattern-recognition approach for electrical engineering with the Massachusetts Insti-
driving skill characterization,” IEEE Trans. Intell. Transp. Syst., vol. 11, tute of Technology (MIT), Cambridge.
no. 4, pp. 905–916, Dec. 2010. She worked in the undergraduate research op-
[38] C. Cortes and V. Vapnik, “Support vector networks,” Mach. Learn., portunities program (UROP) at the MIT Aerospace
vol. 20, no. 3, pp. 273–297, Sep. 1995. Controls Labs. She contributed to the filtering,
[39] C. M. Bishop, Pattern Recognition and Machine Learning (Information processing, and extraction of vehicle trajectories
Science and Statistics). New York: Springer-Verlag, Aug. 2006. from the CICAS-V real-traffic database.
[40] D. Cutting, J. Kupiec, J. Pedersen, and P. Sibun, “A practical part-of- Ms. Stephens was one of the recipients of the
speech tagger,” in Proc. 3rd Conf. Appl. Nat. Lang. Process., 1992, Anita Borg Google Scholarship in 2010.
pp. 133–140.
[41] J. Bilmes, “A gentle tutorial on the EM algorithm and its application to
parameter estimation for Gaussian mixture and hidden Markov models,”
Int. Comput. Sci. Inst., Berkeley, CA, Tech. Rep. ICSI-TR-97-021, 1997. Jonathan P. How (SM’05) received the B.A.Sc.
[42] Z. R. Doerzaph and V. Neale, “Data acquisition method for developing degree from the University of Toronto, Toronto, ON,
crash avoidance algorithms through innovative roadside data collection,” Canada, in 1987 and the S.M. and Ph.D. degrees in
presented at the Transportation Research Board 89th Annu. Meeting, aeronautics and astronautics from the Massachusetts
Washington, DC, 2010, Paper 10-2762. Institute of Technology (MIT), Cambridge, in 1990
[43] M. Maile and L. Delgrossi, “Cooperative intersection collision avoidance and 1993, respectively.
system for violations (CICAS-V) for avoidance of violation-based inter- He was an Assistant Professor with the Depart-
section crashes,” presented at the Enhanced Safety Vehicles, Stuttgart, ment of Aeronautics and Astronautics, Stanford Uni-
Germany, 2009, Paper 09-0118. versity, Stanford, CA. He is a Richard C. Maclaurin
[44] S. McLaughlin, J. Hankey, and T. Dingus, “A method for evaluating col- Professor with the Department of Aeronautics and
lision avoidance systems using naturalistic driving data,” Accident Anal. Astronautics, MIT. He studied for two years at MIT
Prev., vol. 40, no. 1, pp. 8–16, Jan. 2008. as a postdoctoral associate for the Middeck Active Control Experiment that flew
[45] C. Chang and C. Lin, “LIBSVM: A library for support vector machines,” onboard the Space Shuttle Endeavour, in March 1995. His current research
ACM Trans. Intell. Syst. Technol. (TIST), vol. 2, no. 3, p. 27, 2011. interests include robust coordination and control of autonomous vehicles in
[46] M. Dunham and K. Murphy, PMTK3—Probabilistic modeling toolkit dynamic uncertain environments.
for matlab/octave, ver. 3. [Online]. Available: http://code.google.com/p/ Dr. How received of the 2002 Institute of Navigation Burka Award and is an
pmtk3/ Associate Fellow of American Institute of Aeronautics and Astronautics.