US0073 89178B2

(12) United States Patent (10) Patent N0.: US 7,389,178 B2

Raz et a]. (45) Date of Patent: Jun. 17, 2008

(54) SYSTEM AND METHOD FOR VEHICLE 4,671,111 A 6/1987 Lemelson

DRIVER BEHAVIOR ANALYSIS AND 5,270,708 A 12/1993 Kamishima et a1.
EVALUATION 5,546,305 A 8/1996 Kondo
5,570,087 A 10/1996 Lemelson
(75) Inventors: Ofer Raz, Moshav Bnay a IL ; Hod 5,805,079 A 9/1998 Lemelson
Fleishman, Jerusalem (IL); Itamar 6,060,989 A 5/2000 Gehlot
Mulchadsky> Tel'Avl" (IL) 6,438,472 B1 8/2002 Tano et a1.
. _ . . . 2002/0128751 A1 * 9/2002 Engstrom et a1. ............ .. 701/1

(73) Asslgnee' grrizlhrggsgvmg Technologles Ltd" 2007/0001831 A1* 1/2007 Raz e161. ................. .. 340/439

(*) Notice: Subject to any disclaimer, the term of this

patent is extended or adjusted under 35 * Cited b examiner
U.S.C. 154(1)) by 698 days. y
Primary Examiner4Gertrude A. Jeanglaude
(21) APP1~ NOJ 10/894-11345 (74) Attorney, Agent, or FirmiBroWdy and Neimark
(22) Filed: Jul. 20, 2004 (57) ABSTRACT
(65) Prior Publication Data
Us 2005/0131597 A1 Jun 16, 2005 A system and method for analyzing and evaluating the
performance and attitude of a motor vehicle driver. A raW
Related US, Application Data data stream from a set of vehicle sensors is ?ltered to
_ _ _ _ eliminate extraneous noise, and then parsed to convert the
(60) 3021330311211 apphcanon NO 60/528522 ?led on Dec stream into a string of driving event primitives. The string of
' driving events is then processed by a pattern-recognition
system to derive a sequence of higher-level driving maneu
(51) Int. Cl. . . . . .
vers. Dr1v1ng maneuvers mclude such fam1l1ar procedures as
G01C 21/00 (2006.01) . . . .
52 U 5 Cl 701/200_ 70109 70184 lane changmg, passmg, and turn and brake. Dr1v1ng events
( ) ' ' ' """""""""""" " 701/36_701/56_340/903 and maneuvers are quanti?ed by parameters developed from
_ _ _ the sensor data. The parameters and timing of the maneuvers
(58) Field of Classi?cation Search .................. .. 701/ 1, can be analyzed to determine Skin and attitude factors for
70109, 31 3334 36> 56> 301; 340/438 evaluating the drivers abilities and safety ratings. The
_ _ 349/439 903 rendering of the data into common driving-related concepts
See apphcanon ?le for Complete Search hlstory' alloWs more accurate and meaningful analysis and evalua
(56) References Cited tion1 than is possible With ordinary statistical threshold-based
ana ys1s.
U.S. PATENT DOCUMENTS
4,500,868 A 2/1985 Tokitsu et a1. 15 Claims, 14 Drawing Sheets

Sensors
K10] 111
103 05 07 109

@Q Other
sensors

[101 (101 RAW DATA STREAM

Event Handler 2M 1 Low-Pass Filter Events Stack and
Event Library |-| Event Detector Event Extractor
+ I \
20: DRN'IG EVENT STRIG 105

219
208 213
Km :15
l i + 1
Detector aneuver Library Pattern Recognition Unit

pi Maneuver Classi?er HSkillAssessorHittitude Assegl \

RMNG SKILL ATTl'l'lDE

. . . EMT ASSESSMENT ASSESSMENT
II

235 1:7 239

U.S. Patent Jun. 17, 2008 Sheet 1 0f 14 US 7,389,178 B2

/101
Sensors K111
(103 (105 (107 (109 other
RPM Speed Acceleration Location sensors

RAW DATA STREAM

113/\
Analysis and Evaluation Unit
115 Continuous Processing Unit
Threshold \_/
Settings 121
+ 119
. . . Statistical
A Dlscrimlnator
117 > Unit J

125 STATISTICALLY-PROCESSED DATA

REPORT /

k
f
SESSION SESSION ROAD-SPECIFI
0 0 0 n AVERAGE SESSION

127 129 131

no. 1 (PRIOR ART)

U.S. Patent Jun. 17, 2008 Sheet 2 0f 14 US 7,389,178 B2

[101
Sensors /"111
(103 (105 (107 (109 other
RPM Speed Acceleration Location sensors

[201 {Z02 RAWDATA STREAM

Event Handler 207 Low-Pass Filter Events Stack and
Event Library - Event Detector Event Extractor
A ) \
203/ DRMNG EVENT STRlNG\205

209
219
208 213 211
l / f /'2l5
Maneuver\ _ i i _
Detector Maneuver Library Pattern Recognition Unlt_
/ Maneuver Classifier Skill Assessor Attitude Assessor
\
L217 % DRIVING MANEUVER SEQUENCE {E \221
223
Anomaly Detector v/ 225/ ANALYZER
DRMNG ANALYSISAND
INCONSISTENCIES EVALUATIONS

227 REPORT!
NOTIFICATION!
ALARM 229

DRMNG SKILL ATTITUDE

, , , SrrUAT'o ASSESSMENT ASSESSMENT
n

231 233 235 237 239

FIG. 2
U.S. Patent Jun. 17, 2008 Sheet 3 0f 14 US 7,389,178 B2

307 305
301
\ \

303
\

309

FIG. 3
U.S. Patent Jun. 17, 2008 Sheet 4 0f 14 US 7,389,178 B2

401

Filtered Data
as
N
25
II
15
IO
5
0
6
40
45
an

403

FIG. 4
U.S. Patent Jun. 17, 2008 Sheet 5 0f 14 US 7,389,178 B2

403

525 527 529 533 535 539 543

i
i
2
.3

521 523 531 537 541

R, ' F' A ,1
w )1 I
Start: s) End.- 8 > Max: M Min: L Cross: C L. Max: j 1.. Flat: Flat: P

501 503 505 507 509 511 513 515

FIG. 5
U.S. Patent Jun. 17, 2008 Sheet 6 0f 14 US 7,389,178 B2

601

iX
/
Y Time

603

FIG. 6
U.S. Patent Jun. 17, 2008 Sheet 7 0f 14 US 7,389,178 B2

701

703 707

705

FIG. 7
U.S. Patent Jun. 17, 2008 Sheet 8 0f 14 US 7,389,178 B2

801

807 809
803

/ 813

FIG. 8
U.S. Patent Jun. 17, 2008 Sheet 9 0f 14 US 7,389,178 B2

901

909 913

FIG. 9
U.S. Patent Jun. 17, 2008 Sheet 10 0f 14 US 7,389,178 B2

Begin 1005 1003 Done

1001/ \ b
A
-> Sx > M): -> Ex - Accelerate
1007
LX+EX - Braking
1009

l>Sx->Mx>-My>-D<+Ey - Turn and Accelerate

\1011

FIG. 10
U.S. Patent Jun. 17, 2008 Sheet 11 0f 14 US 7,389,178 B2

I
1101 1103 START
+
V Raw Sensor
Data Stream
7 \_ Filter
Sensor Data Stream
1105
+
\_ Detect Events in
1107 1109 Filtered Data Strean

\7/ Event String / \ Generate*Event Strir

1111
+
\_ Match Event String
Patterns to Maneuve "
1113 1115
+
\/
1117
Maneuver
Sequence
/ K Generate
1119
Maneuve
Sequence

\7/ Skill Rating \ Assess *Driver Skill

1121 1123 +
\_/ Attitude Rating / \\/ Assess Driver Attitud
1125 1127 +
- Detect Driving
\_/ Inconsistency Set / p Anomalies
1131
\_, Initiate Alert

> Analyze and Evaluat

1135 +
\_, Issue Reports

1139
\_, Initiate Alert

DONE
4

FIG. 11
U.S. Patent Jun. 17, 2008 Sheet 12 0f 14 US 7,389,178 B2

1201
\ Maneuver 1213

1207

1209
1203 1205

Poorly-Skilled ) K Highly-Skilled
Maneuver Maneuver

213

\ Maneuver Library

FIG. 12
U.S. Patent Jun. 17, 2008 Sheet 13 0f 14 US 7,389,178 B2

1201

\ Maneuver 1313

1307

1303 1305

Safely-Executed ) KDangerously- Executed

Maneuver Maneuver

T T
213

\ Maneuver Library

FIG. 13
U.S. Patent Jun. 17, 2008 Sheet 14 0f 14 US 7,389,178 B2

1407

Threshold
1409

DRIVING
INCONSISTENCIES

1405

1401 1403

Maneuver Characterlstlc Maneuver

209
Database

FIG. 14
 US 7,389,178 B2
1 2
SYSTEM AND METHOD FOR VEHICLE
DRIVER BEHAVIOR ANALYSIS AND
EVALUATION
The present application claims bene?t of US. Provisional
Patent Application No. 60/528,522 ?led Dec. 11, 2003.
FIELD OF THE INVENTION
The present invention relates to vehicle monitoring sys
tems, and methods, and, more particularly, to systems and
methods for monitoring and evaluating vehicle driver behav
ior.
BACKGROUND OF THE INVENTION
There are recognized bene?ts in having systems and
methods to monitor the operation of vehicles, for capturing
real-time data pertaining to driving activity and patterns
thereof. Such systems and methods facilitate the collection
of qualitative and quantitative information related to the
contributing causes of vehicle incidents, such as accidents;
and alloW objective driver evaluation to determine the
quality of driving practices. The potential bene?ts include
preventing or reducing vehicle accidents and vehicle abuse;
and reducing vehicle operating, maintenance, and replace
ment costs. The social value of such devices and systems is
universal, in reducing the impact of vehicle accidents. The
economic value is especially signi?cant for commercial and
institutional vehicle ?eets, as Well as for general insurance
and risk management.
There exists a large and growing market for vehicle
monitoring systems that take advantage of neW technologi
cal advances. These systems vary in features and function
ality and exhibit considerable scope in their approach to the
overall problem. Some focus on location and logistics,
others on engine diagnostics and fuel consumption, Whereas
others concentrate on safety management.
For example, US. Pat. No. 4,500,868 to Tokitsu et al.
(herein denoted as Tokitsu) is intended as an adjunct in
driving instruction. By monitoring a variety of sensors (such
as engine speed, vehicle velocity, selected transmission gear,
and so forth), a system according to Tokitsu is able to
determine if certain predetermined condition thresholds are
exceeded, and, if so, to signal an alarm to alert the driver.
Alarms are also recorded for later revieW and analysis. In
some cases, a simple system such as Tokitsu can be valuable.
For example, if the driver Were to strongly depress the
accelerator pedal, the resulting acceleration could exceed a
predetermined threshold and sound an alarm, cautioning the
driver to reduce the acceleration. If the driver Were prone to
such behavior, the records created by Tokitsus system
Would indicate this. On the other hand, Tokitsus system is
of limited value under other conditions. For example, if the
driver Were to suddenly apply the vehicle brakes With great
force, the resulting deceleration could exceed a predeter
mined threshold, and thereby signal an alarm and be
recorded. Although the records of such behavior could be
valuable, such strong braking is usually done under emer
gency conditions Where the driver is already aWare of the
emergency, and Where an alarm Would be super?uous (and
hence of little or no value), or perhaps distracting (and hence
of dubious value or even detrimental).
US. Pat. No. 4,671,111 to Lemelson (herein denoted as
Lemelson 111) teaches the use of accelerometers and data
recording/transmitting equipment for obtaining and analyZ
ing vehicle acceleration and deceleration. Although Lemel
son 111 presents this in the context of analyZing vehicle
performance, hoWever, there is no detailed discussion of
precisely hoW an analysis of the resulting data Would be
done, nor hoW meaningful information could be obtained
thereby. In related US. Pat. No. 5,570,087 also to Lemelson
(herein denoted as Lemelson 087) the analyZed vehicular
motion is expressed in coded representations Which are
stored in computer memory. As With Lemelson 111, Which
does not describe hoW raW data is analyZed to determine
driving behavior parameters, Lemelson 087 does not
describe hoW coded representations of raW data or driving
behavior parameters Would be created or utiliZed. It is
further noted that US. Pat. No. 5,805,079 to Lemelson
(herein denoted as Lemelson 079) is a continuation of
Lemelson 087 and contains no neW or additional descriptive
material.
US. Pat. No. 5,270,708 to Kamishima (herein denoted as
Kamishima) discloses a system that detects a vehicles
20
position and orientation, turning, and speed, and coupled
With a database of past accidents at the present location,
determines Whether the present vehicles driving conditions
are similar to those of a past accident, and if so, alerts the
driver. If, for example, the current vehicle speed on a
25
particular road exceeds the (stored) speed limit at that point
of the road, the driver could be alerted. Moreover, if exces
sive speed on that particular area is knoWn to have been
responsible for many accidents, the system could notify the
driver of this. The usefulness of such a system, hoWever,
30
depends critically on having a base of previous data and
being able to associate the present driving conditions With
the stored information. The Kamishima system, in particu
lar, does not analyZe driving behavior in general, nor draW
any general conclusions about the drivers patterns in a
35
location-independent manner.
US. Pat. No. 5,546,305 to Kondo (herein denoted as
Kondo) performs an analysis on raW vehicle speed and
acceleration, engine rotation, and braking data by time
dilferentiating the raW data and applying threshold tests.
40 Although such an analysis can often distinguish betWeen
good driving behavior and erratic or dangerous driving
behavior (via a driving roughness analysis), time-differ
entiation and threshold detection cannot by itself classify
raW data streams into the familiar patterns that are normally
45 associated With driving. Providing a count of the number of
times a driver exceeded a speed threshold, for example, may
be indicative of unsafe driving, but such a count results in
only a vague sense of the drivers patterns. On the other
hand, a context-sensitive report that indicates the driver
50 repeatedly applies the brake during turns Would be far more
revealing of a potentially-dangerous driving pattern. Unfor
tunately, hoWever, the analysis performed by Kondo, Which
is typical of the prior art analysis techniques, is incapable of
providing such context-sensitive information. (See Limita
55 tions of the Prior Art beloW.)
US. Pat. No. 6,060,989 to Gehlot (herein denoted as
Gehlot) describes a system of sensors Within a vehicle for
determining physical impairments that Would interfere With
a drivers ability to safely control a vehicle. Speci?c physi
60 cal impairments illustrated include intoxication, fatigue and
droWsiness, or medicinal side-effects. In Gehlots system,
sensors monitor the person of the driver directly, rather than
the vehicle. Although this is a useful approach in the case of
physical impairments (such as those listed above), Gehlots
65 system is ineffective in the case of a driver Who is simply
unskilled or Who is driving recklessly, and is moreover
incapable of evaluating a drivers normal driving patterns.
 US 7,389,178 B2
3 4
Us. Pat. No. 6,438,472 to Tano, et al. (herein denoted as
Tano) describes a system Which analyzes raW driving data
(such as speed and acceleration data) in a statistical fashion
to obtain statistical aggregates that can be used to evaluate
driver performance. Unsatisfactory driver behavior is deter
mined When certain prede?ned threshold values are
exceeded. A driver Whose behavior exceeds a statistical
threshold from What is considered safe driving, can be
deemed a dangerous driver. Thresholds can be applied to
various statistical measures, such as standard deviation.
Because Tano relies on statistical aggregates and thresholds
Which are acknowledged to vary according to road location
and characteristics, hoWever, a system according to Tano has
limited ability to evaluate driver performance independent
of the statistical pro?les and thresholds. In particular, the
statistical characterization of a drivers performance is gen
erally not expressible in terms of familiar driving patterns.
For example, a driver may have a statistical pro?le that
exceeds a particular lateral acceleration threshold, and the
driver may therefore be classi?ed as a dangerous driver.
But What driving pattern is responsible for excessive lateral
acceleration? Is it because this driver tends to take curves too
fast? Or is it because he tends to change lanes rapidly While
Weaving in and out of traf?c? Both are possibly dangerous
patterns, but a purely threshold-oriented statistical analysis,
such as presented in Tano, may be incapable of discrimi
nating betWeen these, and therefore cannot attribute the
resulting statistical pro?le to speci?c patterns of driving. As
noted for Kondos analysis (above), Tanos statistical analy
sis is also incapable of providing information in terms of
familiar driving patterns.
In addition to the above issued patents, there are several
commercial products currently available for monitoring
vehicle driving behavior. The Mastertrak system by
Vetronix Corporation of Santa Barbara, Calif. is intended as
a ?eet management system Which provides an optional
safety module. This feature, hoWever, addresses only
vehicle speed and safety belt use, and is not capable of
analyZing driver behavior patterns. The system manufac
tured by SmartDriver of Houston, Tex. monitors vehicle
speed, accelerator throttle position, engine RPM, and can
detect, count, and report on the exceeding of thresholds for
these variables. Unfortunately, hoWever, there are various
driving patterns Which cannot be classi?ed on the basis of
thresholds, and Which are nevertheless pertinent to detecting
questionable or unsafe driving behavior. For example, it is
generally acknowledged that driving too sloWly on certain
roads can be haZardous, and for this reason there are often
minimum speed limits. Driving beloW a minimum speed,
hoWever, is not readily detectable by a system such as
SmartDriver, because introducing a loW-speed threshold
results in such a large number of false reports (When the
vehicle is driven sloWly in an appropriate location) that
collecting such data is not normally meaningful.
Limitations of the Prior Art
Collecting raW physical data on vehicle operation through
a multiplicity of sensors usually results in a very large
quantity of data Which is cumbersome to store and handle,
and impractical to analyZe and evaluate. For this reason, any
automated system or method of driver behavior analysis and
evaluation must employ some abstraction mechanism to
reduce the data to a manageable siZe and in a meaningful
Way.
For the prior art, as exempli?ed by the speci?c instances
cited above, this is done through statistical processing and
the use of predetermined thresholds, supplemented in some
cases by limited continuous pre-processing (e.g., time-dif
ferentiation), optionally correlated in some cases With avail
able history or other data on the location Where the driving
is being done. As a result, prior art systems and methods are
generally limited to providing aggregate and statistically
processed overvieWs of driver performance. This is
expressed succinctly in Lemelson 111: The computer
analysis may determine the manner in Which the vehicle is
driven, either during a speci?c time interval or a number of
time intervals or over a longer period of time Wherein
averaging is employed to determine the general performance
or use of the vehicle(column 1 lines 21-26). That is, prior
art analysis and evaluation is based on overall performance
during a particular driving session, or is based on statistical
averages over a number of different sessions. In limited
cases, the analysis and evaluation can be made With regard
to a particular road or road segment, through the application
of GPS locating.
FIG. 1 illustrates the general prior art analysis and evalu
20 ation approach. Atypical set of sensors 101 has a tachometer
103, a speedometer 105, one or more accelerometers 107, a
GPS receiver 109, and optional additional sensors 111. In the
case of accelerometers, it is understood that an accelerom
eter is typically operative to monitoring the acceleration
25 along one particular speci?ed vehicle axis, and outputs a raW
data stream corresponding to the vehicles acceleration
along that axis. Typically, the tWo main axes of vehicle
acceleration that are of interest are the longitudinal vehicle
axisithe axis substantially in the direction of the vehicles
30 principal motion (forWard and reverse); and the trans
verse (lateral) vehicle axisithe substantially horiZontal axis
substantially orthogonal to the vehicles principal motion
(side-to-side). An accelerometer Which is capable of
monitoring multiple independent vector accelerations along
35 more than a single axis (a multi-axis accelerometer) is
herein considered as, and is denoted as, a plurality of
accelerometers, Wherein each accelerometer of the plurality
is capable of monitoring the acceleration along only a single
axis. Additional sensors can include sensors for driver
40 braking pressure, accelerator pressure, steering Wheel con
trol, handbrake, turn signals, and transmission or gearbox
control, clutch (if any), and the like. Some of the sensors,
such as tachometer 103 and speedometer 105 may simply
have an analog signal output Which represents the magnitude
45 of the quantity. Other sensors, such as a transmission or
gearbox control sensor may have a digital output Which
indicates Which gear has been selected. More complex
output Would come from GPS receiver 109, according to the
formatting standards of the manufacturer or industry. Other
50 sensors can include a real-time clock, a directional device
such as a compass, one or more inclinometers, temperature
sensors, precipitation sensors, available light sensors, and so
forth, to gauge actual road conditions and other driving
factors. Digital sensor output is also possible, Where sup
55 ported. The output of sensor set 101 is a stream of raW data,
in analog and/or digital form.
Sensor outputs are input into an analysis and evaluation
unit 113, Which has threshold settings 115 and a threshold
discriminator 117. A statistical unit 119 provides report
60 summaries, and an optional continuous processing unit 121
may be included to preprocess the raW data. The output of
analysis and evaluation unit 113 is statistically-processed
data.
A report/noti?cation/alarm 123 is output With the results
65 of the statistical analysis, and may contain analysis and
evaluations of one or more of the folloWing: an emergency
alert 125, a driving session 1 statistics report 127, a driving
 US 7,389,l78 B2
5 6
session 2 statistics report 129, etc., and a driving session n
statistics report 131, a driving session average statistics
report 133, and a road-speci?c driving session statistics
report 135.
These reports may be useful in analyzing and evaluating
driver behavior, skill, and attitude, but the use of statistics
based predominantly on thresholds or on localiZation of the
driving, and the aggregation over entire driving sessions or
groups of driving sessions also result in the loss of much
meaningful information.
In particular, the details of the drivers behavior in speci?c
driving situations are not available. Familiar driving situa
tions, such as passing, lane changing, tra?ic blending, mak
ing turns, handling intersections, handling olf- and on
ramps, driving in heavy stop- and-go traf?c, and so forth,
introduce important driving considerations. It is evident that
the aggregate statistics for a given driver in a given driving
session depend critically on the distribution and mix of these
situations during that given session.
For example, the same driver, driving in a consistent
manner but handling different driving situations may exhibit
completely different driving statistics. One of the key ben
e?ts of monitoring driving behavior is the ability to deter
mine a drivers consistency, because this is an important
indicator of that drivers predictability, and therefore of the
safety of that drivers performance. If the driver begins to
deviate signi?cantly from an established driving pro?le, this
can be a valuable advance warning of an unsafe condition.
Perhaps the driver is fatigued, distracted, or upset, and
thereby poses a haZard Which consistency analysis can
detect. It is also possible that the driver has been misiden
ti?ed and is not the person thought to be driving the vehicle.
Unfortunately, hoWever, statistically aggregating data, as is
done in the prior art, does not permit a meaningful consis
tency analysis, because such an analysis depends on the
particular driving situations Which are encountered, and
prior art analysis completely ignores the speci?cs of those
driving situations.
Atypical prior art report presents information such as: the
number of times a set speed limit Was exceeded; the maxi
mum speed; the number of times a set RPM limit Was
exceeded; the maximum lateral acceleration or braking
deceleration; and so forth. Such information may be char
acteristic of the drivers habits, but it Would be much better
to have a report that is based on familiar driving situations,
maneuvers, and patternsifor example, by revealing that the
driver has a habit of accelerating during turns, or makes
frequent and rapid high-speed lane changes.
As another example of the limitations of the prior art, a
neW and relatively inexperienced driver might drive very
cautiously and thereby have very safe overall statistics,
but might lack skills for handling certain common but more
challenging driving situations. An experienced driver, hoW
ever, might exhibit What appear to be more dangerous
overall statistics, but might be able to handle those chal
lenging driving situations much better and more safely than
the neW driver. Prior art analysis systems and methods,
hoWever, might erroneously deduce that the more experi
enced driver poses the greater haZard, Whereas in reality it
is the apparently safer driver Who should be scrutinized
more carefully.
There is thus a need for, and it Would be highly advan
tageous to have, a method and system for analyZing a raW
vehicle data stream to determine the corresponding sequence
of behavior and characteristics of the vehicles driver in the
context of familiar driving situations, and for expressing
driving behavior and characteristics in terms of familiar
driving patterns and maneuvers. This goal is met by the
present invention.
SUMMARY OF THE INVENTION
The present invention is of a system and method for
analyZing and evaluating a raW data stream related to the
operation of a vehicle for the purpose of classifying and
rating the performance of the vehicles driver. Unlike prior
art systems and methods, embodiments of the present inven
tion are not restricted to performing statistical and threshold
analysis and evaluation of the drivers skills and behavior,
but can express an analysis and evaluation in terms of
familiar driving patterns and maneuvers, thereby yielding
analyses and evaluations Which contain more information
and Which are more readily put to use.
According to embodiments of the present invention, the
raW data stream from the vehicle sensors is progressively
20
analyZed to obtain descriptors of the driving operations
Which are less and less data and more and more expressive
of familiar driving operations and situations. An objective of
the present invention is to identify the context in Which each
event takes place. For example, braking suddenly is de?ned
25
as an event, and the context of such an event may be the
making of a turn Which Was entered at too high a speed.
It is thus an objective of the present invention to identify
events in the context of driving situations. Whereas prior art
solutions perform statistical analysis according to threshold
levels of measurable variables (such as counting the number
of times a driver exceeds a particular speed), it is a goal of
the present invention to recogniZe common patterns of
driving situations (such as lane changing) and to associate
other events With a context (such as increasing speed during
35
a lane change) Which may represent unsafe driving behavior.
It is an objective of the present invention to facilitate the
classi?cation of a drivers skill on the basis of sensor
measurements of the driven vehicle. It is also an objective of
40
the present invention to facilitate the classi?cation of a
drivers attitude on the basis of sensor measurements of the
driven vehicle, Where the term attitude herein denotes the
drivers approach toWard driving and the tendency of the
driver to take risks. Categories include, but are not limited
45 to: safe (or normal); aggressive (or risky); thrill
seeking; abusive; and dangerous. In an embodiment of
the present invention, aggressive or dangerous behavior is
logged as an event.
It is moreover an objective of the present invention to
50 enable the making of quantitative and qualitative compari
sons betWeen a current drivers behavior and a previous
pro?le of the same driver, independent of the particular
details of the driving sessions involved, by qualifying and
quantifying the drivers behavior When performing common
55 driving maneuvers in common driving situations.
Therefore, according to the present invention there is
provided a system for analyZing and evaluating the perfor
mance and behavior of the driver of a vehicle, the system
including: (a) a vehicle sensor operative to monitor the state
60 of the vehicle and to output a raW data stream corresponding
thereto; (b) a driving event handler operative to detect
driving events based on the raW data stream and to output a
driving event string corresponding thereto, the driving event
string containing at least one driving event symbol corre
65 sponding to a driving event; and (c) a maneuver detector
operative to recogniZe patterns of driving maneuvers and to
construct and output a driving maneuver sequence corre
 US 7,389,178 B2
7 8
sponding thereto, the driving maneuver sequence containing
at least one driving maneuver.
In addition, according to the present invention there is
also provided a method for analyzing and evaluating the
performance and behavior of the driver of a vehicle, based
on a raW data stream from a set of sensors operative to
monitor the state of the vehicle, the method including: (a)
detecting driving events represented by the raW data stream,
and generating a driving event string therefrom; and (b)
matching patterns in the driving event string to detect
driving maneuvers therein, and generating a driving maneu
ver sequence therefrom.
BRIEF DESCRIPTION OF THE DRAWINGS
The invention is herein described, by Way of example
only, With reference to the accompanying draWings,
Wherein:
FIG. 1 conceptually illustrates prior art analysis and
evaluation of vehicle driving data.
FIG. 2 is a block diagram of a system according to an
embodiment of the present invention.
FIG. 3 is an example of a graph of a raW data stream from
multiple vehicle accelerometers.
FIG. 4 is an example of the ?ltering of the raW data stream
to remove noise, according to the present invention.
FIG. 5 is an example of parsing a ?ltered data stream to
derive a string of driving events, according to the present
invention.
FIG. 6 shoWs the data and event string analysis for a lane
change driving maneuver, according to the present inven
tion.
FIG. 7 shoWs the data and event string analysis for a
tum driving maneuver, according to the present invention.
FIG. 8 shoWs the data and event string analysis for a
braking Within turn driving maneuver, according to the
present invention.
FIG. 9 shoWs the data and event string analysis for an
accelerate Within turn driving maneuver, according to the
present invention.
FIG. 10 shoWs a non-limiting illustrative example of
transitions of a ?nite state machine for identifying driving
maneuvers, according to an embodiment of the present
invention.
FIG. 11 is a ?owchart of a method for analyZing and
evaluating vehicle driver performance according to an
embodiment of the present invention.
FIG. 12 is a conceptual block diagram of an arrangement
for assessing driver skill according to an embodiment of the
present invention.
FIG. 13 is a conceptual block diagram of an arrangement
for assessing driver attitude according to an embodiment of
the present invention.
FIG. 14 is a conceptual block diagram of an arrangement
for determining Whether there is a signi?cant anomaly in the
current drivers behavior and/or performance according to
an embodiment of the present invention.
DESCRIPTION OF THE PREFERRED
EMBODIMENTS
The principles and operation of a system and method
according to the present invention may be understood With
reference to the draWings and the accompanying description.
System and Data Progression
FIG. 2 illustrates a system according to an embodiment of
the present invention. Sensor set 101 is comparable to that
of the prior art system illustrated in FIG. 1, for monitoring
states of the vehicle, and having an output in the form of a
raW data stream. The raW data is input into a driving event
handler 201, Which contains a loW-pass ?lter 202, a driving
event detector 203, a driving events stack and driving event
extractor 205 for storing and managing the driving events,
and a driving event library 207, Which obtains speci?c data
from a database 209.
According to the present invention, driving events are
primitive driving operations that characteriZe basic moves
of driving, as explained and illustrated in detail beloW.
Driving event handler 201 performs a basic analysis on the
raW data from sensor set 101, and outputs a string of driving
events corresponding to the raW data stream. A driving event
string is represented as a time-ordered non-empty set of
driving event symbols arranged in order of their respective
20 occurrences. Driving event detector 203 performs a best-?t
comparison of the ?ltered sensor data stream With event
types from event library 207, such as by using the Well
knoWn sliding WindoW technique over the data stream. A
real-time clock 208 provides a reference time input to the
25 system, illustrated here for a non-limiting embodiment of
the present invention as input to driving event handler 201.
Furthermore, according to embodiments of the present
invention, a driving event is characterized by a symbol that
qualitatively identi?es the basic driving operation, and may
30 be associated With one or more numerical parameters Which
quantify the driving event. These parameters may be derived
from scaling and offset factors used in making the best-?t
comparison against events from event library 207, as
described above. For example, the scaling of the time axis
35 and the scaling of the variable value axis Which produce the
best ?t of the selected segment of the input data stream to the
model of the event in event library 207 can be used as
numerical parameters (in most cases, one or more of these
numerical parameters are related to the beginning and end
40 times of the driving event). If close ?ts can be obtained
betWeen the string of driving events and the input data
stream, the event string (including the event symbols and
associated parameter set) can replace the original data
stream, thereby greatly compressing the data and providing
45 an intelligent analysis thereof.
As a non-limiting example, a simple event is to start the
vehicle moving forWard from a stopped position (the start
event). A numerical parameter for this event is the magni
tude of the acceleration. A generaliZed version of this event
50 is to increase the speed of a moving vehicle (the accelerate
event). Another simple event is to sloW the vehicle to a halt
from a moving condition (the stop event). Other events are
of like simplicity. In place of a continuous stream of data
from the sensors, Which is the input to event handler 201, the
55 output driving event string is a sequence of basic driving
events as explained above.
The driving event string is then input into a driving
maneuver detector 211. According to the present invention,
a driving maneuver is a combination of driving events Which
60 are encountered as a familiar pattern in normal driving. A
lane change, for example, is a driving maneuver that, in
the simplest case, may be represented by a combination of
a lateral acceleration folloWed by a lateral deceleration
during a period of forWard motion. A lane change during a
65 turn is more involved, but can be similarly represented by a
combination of driving events. As in the case of the driving
events themselves, driving maneuvers can contain one or

more numerical parameters, Which are related to the numeri
cal parameters of the driving events Which make up the
driving maneuver.
A driving maneuver sequence is a time-ordered non
empty set of driving maneuvers arranged according to the
respective times of their occurrence. Returning to FIG. 2, it
is seen that in order to derive driving a sequence of driving
maneuvers from a string of driving events, maneuver detec
tor 211 contains a maneuver library 213 fed from database
209, a pattern recognition unit 215 to recogniZe patterns of
driving maneuvers to identify clusters of driving events
Which make up driving maneuvers, and a maneuver classi
?er 217 to construct a reasonable driving maneuver
sequence output corresponding to the input driving event
string. Exemplary non-limiting patterns include sequences
of events such as accelerating out of stops and changing
lanes While speeding or approaching turns too fast. By
comparing the timing and other quantities of the driving
maneuver With those of knoWn skillful drivers, a skill
assessor 219 can develop and assign a skill rating for the
current drivers handling of the driving maneuver. Further
more, by analyZing the magnitude of certain key parameters
(such as those related to acceleration and deceleration during
the maneuver), an attitude assessor 221 can develop and
assign an attitude rating to the current drivers execution of
the driving maneuver. Moreover, each maneuver is assigned
a Weighting driving risk coe?icient for developing and
assigning an aggregate attitude rating for the current driver.
The folloWing Table 1 includes non-limiting examples of
some common driving maneuvers, their common meaning
in a driving context, and their suggested driving risk coef
?cients. In a non-limiting example, coef?cients range from
1 to 10, With 10 representing the most dangerous driving
maneuvers. Risk coef?cients, of course, are subjective, and
according to an embodiment of the present invention may be
rede?ned to suit empirical evidence.
TABLE 1
Examples of Driving Maneuvers and Driving Risk Coefficients
Driving Maneuver Coefficient
Accelerate
increase vehicle speed
Accelerate before Turn
increase vehicle speed prior to a turn
Accelerate during Lane Change
increase vehicle speed While moving to a different
travel lane
Accelerate into Turn
Increase vehicle speed While initiating a turn
Accelerate into Turn out of Stop
start moving vehicle While initiating a turn from a
stopped position
Accelerate out of Stop
start moving vehicle from a stopped position
Accelerate out of Turn
increase vehicle speed While completing a turn
Accelerate While Passing
increase vehicle speed While overtaking and
bypassing a leading vehicle When initially traveling
in the same travel lane
Braking
applying vehicle brakes to reduce speed
Braking after Turn
applying vehicle brakes to reduce speed after
completing a turn
Braking before Turn
applying vehicle brakes to reduce speed before
beginning a turn
US 7,389,178 B2
10
TABLE 1-continued
Examples of Driving Maneuvers and Driving Risk Coefficients
Driving Maneuver Coefficient
Braking into Stop 3
applying vehicle brakes to reduce speed and coming
to a stopped position
Braking out of Turn 7
applying vehicle brakes to reduce speed While
completing a turn
Braking Within Turn 8
applying vehicle brakes to reduce speed during a
turn
Failed Lane Change
aborting an attempted move to a different travel lane
Failed Passing
aborting an attempt to overtake and bypass a leading
vehicle When initially traveling in the same travel
lane
Lane Change 4
moving into a different travel lane
20
Lane Change and Braking 8
moving into a different travel lane and then applying
vehicle brakes to reduce speed
Passing 4
overtaking and bypassing a leading vehicle When
initially traveling in the same travel lane
25 Passing and Braking 8
overtaking and passing a leading vehicle When
initially traveling in the same travel lane and then
applying vehicle brakes to reduce speed
Turn 3
substantially changing the vehicle travel direction
30 Turn and Accelerate 4
substantially changing the vehicle travel direction
and then increasing vehicle speed
U-Turn 5
substantially reversing the vehicle travel direction
35
FolloWing processing by driving maneuver detector 211,
a driving anomaly detector 223 checks the output driving
maneuvers for inconsistencies in the driving pro?le of the
driver. A pro?le or set of pro?les for a driver can be
40
maintained in database 209 for comparison With the drivers
current behavior. A set of pro?les for various maneuvers can
be maintained so that Whatever the current driving maneuver
happens to be, a comparison can be made With a similar
recorded maneuver. If there is a substantial discrepancy
3 betWeen the current driving maneuvers and stored pro?les
45
6
for the driver Which are used as reference, the driving
inconsistencies can be reported to an emergency alert 227
5 for folloW-up checking or investigation. As previously
noted, a signi?cant discrepancy or inconsistency may indi
5
cate an unsafe condition.
50
The sequence of driving maneuvers that is output by
6 driving maneuver detector 211 also goes to an analyZer 225,
Which outputs analysis and evaluation of the driving behav
ior to a report/noti?cation/alarm 229. As is illustrated in
5
55
FIG. 2, report/noti?cation/alarm 229 can contain informa
4 tion on a driving situation 1 analysis report 231, a driving
situation 2 analysis report 233, etc., and a driving situation
5
n analysis report 235. In addition, by statistically-processing
the driving situation analysis reports, it is possible to pro
60
duce some overall analyses and evaluations, such as a
5 driving skill assessment report 237 and a driving attitude
6
assessment report 239.
Analysis of RaW Data to Obtain a Driving Event String
7
FIG. 3 illustrates an example of raW data from multiple
65 vehicle accelerometers, as plotted in a 3-dimensional form.
An x-axis 301 represents the longitudinal acceleration of the
vehicle (in the direction in Which the vehicle is normally