Naturalistic Driving Studies: Another tool to assess the impact of driver distraction.
description
Transcript of Naturalistic Driving Studies: Another tool to assess the impact of driver distraction.
Naturalistic Driving Studies: Another tool to assess the impact of driver distraction.
Charlie Klauer, Ph.D.
What is ‘Naturalistic’ Driving?• No experimenter present
• Data collected in privately-owned vehicle, preferably
• Instrumentation = unobtrusive– Hidden radar– Face camera behind smoked glass– Other cameras are 1” x 1”
• Preferable to do ‘large-scale’– Collect crash data– Generalizability
What are the advantages of the “Naturalistic” approach?
• More detailed pre-crash/pre-near crash information
• Greater external validity– Information about driver behavior under normal
day-to-day pressures• Rich database to mine
– GES type crash database– Video– Driving performance
Empirical Data Collection
Large-Scale Naturalistic Data
Collection
• Proactive• Provides
important ordinal crash risk info
• Precise knowledge about crash risk
• Information about important circumstances and scenarios that lead to crashes
• Imprecise, relies on unproven safety surrogate
• Experimental situations modify driver behavior
• Reactive• Very limited
pre-crash information
• “Natural” driver behavior in full driving context
• Detailed pre-crash/crash info including driver performance/ behavior, driver error and vehicle kinematics
• Can utilize combination of crash, near crash and other safety surrogate data
Epidemiological Data
Collection
What are the disadvantages of the “Naturalistic” approach?
• Costly and logistically complex• No experimental control over driver experience• Large amounts of data that must be
organized/reduced
100 Car “Naturalistic” Approach
• Data collection in a “naturalistic” setting to obtain crash/pre-crash/near-crash/conflict data as well as distributions of driver performance:– 100 drivers in their own (or leased) cars with
specialized instrumentation, on public roads, as close to unobserved as possible.
– Subjects used instrumented cars for an extended period (up to 13 months) without an experimenter present.
– Subjects were not coached or instructed to perform any specific actions other than drive as they normally would.
– Instrumentation was unobtrusive and inconspicuous to other drivers, but not invisible.
Are the 100 car participants representative of the driving population?
• 241 total drivers; recruited from flyers and newspaper ads
• Ages 18-73• Range of very “safe” to very “unsafe”• Wide range of miles driven• Commonly drove on all road classes• Urban, suburban and a small amount of rural
driving• Both sedans and SUVs
To what degree are the 100 car participants representative of the driving population?
• Possible recruiting bias due to nature of the study
• Extreme age groups missing (i.e., 16-17 and 75+)
• One location (metro DC/NoVA)
• Relatively little rural driving
• Fewer SUVs and light trucks than the national average
To what degree did the presence of the instrumentation affect natural behavior?
• Answer: Immeasurably, after the first hour.• Numerous occurrences of extreme driving across the
sample, including: Impairment, traffic violation, aggressive driving, “road rage.”
• Other “behavioral indicators” were often present.
• Performance data (e.g., reaction time, speed selection) remained largely consistent with prior research and across the 13 months of data collection.
• A few, isolated instances of “showing off” among younger drivers in the presence of other passengers was the only counter-indication.
Naturalistic Data Collection Approach
Highly capable instrumentation (well beyond EDRs)• Five channels of digital, compressed video• Four radar sensors front, rear (for all 100 cars), and
side (for 20 cars)• Machine vision-based lane tracker• Many other sensors: GPS, glare, RF, acceleration,
yaw rate, controls, etc.• Cell phone, wireless internet, or hardwire download• Tie into vehicle network to obtain other sensor
information• Ruggedized, crash tested, all solid state• Crash detection, Fault detection• Remote Access
100 Car Instrumentation Mounted in Trunk
100 Car Face View Camera (behind smoked glass) and and Incident Push Button mounted above Rear View Mirror
100 Car Forward Camera and Glare Sensor Mounted behind the Rear View Mirror
The 100 Car “Event” Database
• Continuous data were initially reduced based on “trigger” signatures from the electronic data that are indicative of the presence of a crash, near crash or conflict/incident event.
• Triggers include:– Radar-based time-to-collision– High lateral acceleration or yaw-rate change– Unplanned lane deviation– High longitudinal decelerations with short time
to collisions– Crashes
The Naturalistic “100 Car” Study
• Once an event was triggered, its validity and severity were determined via a review of 40 seconds of video.
• Some events were triggered by normal driving maneuvers or “false” radar targets (e.g., “flying pass”).
• Severity (i.e., crash, near crash or incident) was determined using the following operational definitions.
Video-Reduced Variables• Pre-Incident maneuver• Crash/Incident type• Precipitating factor• Contributing factor(s)• Evasive maneuver• Roadway/Traffic variables• Weather/Lighting• Driver’s state
Eye glance location Observer rating of drowsiness
• Fault assignment• Crash reconstruction
The Naturalistic “100 Car” Study:Database Statistics
• 42,300 hours of driving data collected/ ~2 M miles of driving• 82 Crashes and collisions
– Defined as any contact between the subject vehicle and another vehicle, fixed object, pedestrian, pedalcyclist, animal.
• 761 Near crashes – Defined as a conflict situation requiring a rapid, severe
evasive maneuver to avoid a crash.• 8295 Incidents
– Conflict requiring an evasive maneuver, but of less magnitude than a near crash.
• 20,000 Normal, baseline driving
Crash/Collision Category Definitions
Collision Category 1 (Police-reported
and/or contains an airbag or injury)Collision Category 2 (Police-reported
with property damage only)
Left Turn Against Path 1 Lane Change 1
Rear-End Struck 2 Left Turn Against Path 1
Run-Off-Road 2 Rear-End Struck 2
Rear-End Strike 5
Run-Off-Road 2
Subtotal 5 Subtotal 11
Collision Category 3 (Non-police-reported, physical contact/property damage)
Collision Category 4 (Non-police-reported, physical contact/no
property damage)
Backing 2 Animal 2
Object 4 Backing 8
Rear-End Strike 6 Object 1
Rear-End Struck 6 Rear-End Strike 6
Run-Off-Road 6 Rear-End Struck 4
Sideswipe 1 Run-Off-Road 20
Subtotal 25 Subtotal 41
Total 82
Uses of Naturalistic Data
• Detailed crash/near crash causation analysis– More pre-crash information than ever before available
• Safety surrogate validation– The relationship between crashes/near
crashes/incidents– The relationship to other surrogates like eye glances
and lane departures• Model development and validation
– Crash benefits estimation– Crash countermeasure assessment
• Modeling example from follow-on project work in progress
100 Car Study Results
The capture of crash/collision events (included minor, non-property-damage contact) provide very valuable information and occur much more frequently (i.e., 5 to 1) than more severe crashes.
This has important implications for future naturalistic driving studies aimed at assessing driver-related crash causation.
100 Car Study Results
2. This study allowed the capture and assessment of near crash events in large numbers. Near crashes provide valuable information as a potential surrogate for crash events and as a tool for the assessment of the factors that contributed to the execution of a successful evasive maneuver.
100 Car Study Results
3. Inattention to the forward roadway, which was operationally defined as including: 1) secondary task distraction, 2) driving-related inattention to the forward roadway (e.g., blind spot checks), 3) moderate to extreme fatigue, and 4) other non-driving-related eye glances, is the primary contributing factor in most crashes and collisions.
100 Car Study Results
4. 80% of all crashes and 65% of all near crashes involved at least one form of driving inattention just prior to (i.e., within 3 seconds) the onset of the conflict.
5. 93% of the conflict with lead vehicle crashes and minor collisions involved looking away. In 86% of the lead vehicle crashes/collisions, the headway at the onset of the event was greater than 2.0 seconds.
Frequency of Events by Inattention Type
0
20
40
60
80
100
120
140
160
Secondary Task
Driving-Related Inattention
Secondary Task + N. S. Eye Glance
Fatigue
Non Specific Eye Glance
Secondary Task + D-R InattentionD-R Inattention + N. S. Eye Glance
Fatigue + N. S. Eye GlanceD-R Inattention + Fatigue
Secondary Task + D-R Inattention + N.
S. Eye Glance
Secondary Task + Fatigue + N. S. Eye
Glance
Types of Inattention
Frequency of Events
Crash
Near-Crash
100 Car Study Results
7.Fatigue contributed to crashes/collisions at much higher rates than is reported using existing crash databases. Fatigue was a contributing factor in 20% of all crashes and 12% of near crashes, while most current database estimates place fatigue-related crashes at approximately 2 to 4% of total crashes.
Frequency of Fatigue-Related Crashes and Near-Crashes Occurring During
Day vs. Night
0
10
20
30
40
50
60
70
Crash Near Crash
Severity
Frequency
Day
Night
100 Car Study Results
9.The rate of inattention-related crash and near crash events decreased dramatically with age, with the rate being as much as four times higher for the 18-20 year old age group relative to some of the older driver groups (i.e., 35 and up).
Distraction Analysis Results
Secondary tasks that are moderately to very complex and driver drowsiness have the highest associated crash risk. Very simple secondary tasks do not appear to have a crash risk that is greater than normal driving.
Secondary Task Complexity Levels
(from Dingus, Antin, Hulse, & Wierwille, 1989). • Complex Secondary Task: Multi-step, multiple
eye glances away from the forward roadway, and/or multiple button presses.
• Moderate Complexity Tasks: At most two glances away from the roadway and/or at most two button presses.
• Simple Secondary Tasks: Zero or one button press and/or one glance away from the forward roadway.
Calculation of Odds Ratio (Relative Risk)
Odds Ratio = (A x D)/(B x C) – A = the number of events where <inattention type> was present without
any other type of inattention
– B = the number of baseline epochs where <inattention type> was present without any other type of inattention
– C = the number of events where < inattention type> was not present or was present but in combination with other types of inattention
– D = the number of baseline epochs where <inattention type> was not present or was present but in combination with other types of inattention
• Interpretation of OR: – Greater than 1.0 = Increased crash risk
– Equal to 1.0 = Risk is same as normal driving
– Less than 1.0 = Decreased crash risk or ‘protective effect’.
Relative Risk Estimates (Odds Ratios) for Crash/Near Crash Inattention Events
Type of Inattention Odds Ratio Lower CL Upper CLComplex Secondary Task 3.1 1.7 5.5
Moderate Secondary Task 2.1 1.6 2.7
Simple Secondary Task 1.2 0.9 1.6
Moderate to Severe Drowsiness (in isolation)
6.2 4.6 8.5
Moderate to Severe Drowsiness (all occurrences)
4.2 3.3 5.5
Driving-related Inattention to the Forward Roadway > 2 s
0.5 0.2 0.8
Driving-related Inattention to the Forward Roadway < 2 s
0.2 0.2 0.3
Non-Specific Eye Glance Away from the Forward Roadway > 2 s
0.9 0.2 3.7
Non-Specific Eye Glance Away from the Forward Roadway < 2 s
0.4 0.2 1.1
Population Attributable Risk (Estimate of the Percentage of Total Crashes where Inattention is a
Contributing Factor)
Type of Inattention Population
Attributable Risk (%)
Lower CL Upper CL
Complex Secondary Task 4.3 4.0 4.6
Moderate Secondary Task
15.2 14.6 15.8
Simple Secondary Task 3.3 2.7 3.9
Moderate to Severe Drowsiness (in isolation)
22.2 21.7 22.7
Moderate to Severe Drowsiness (all occurrences)
24.7 21.1 25.2
Preliminary Results from 100-Car StudyType of Secondary Task Odds Ratio Lower CL Upper CLReaching for a moving object 8.8 2.5 31.2
Insect in vehicle 6.4 0.8 53.1
Looking at external object 3.7 1.1 12.2
Reading 3.4 1.7 6.5
Applying make-up 3.1 1.3 7.9
Dialing hand-held device 2.8 1.6 4.9
Inserting/retrieving CD 2.3 0.3 17.0
Eating 1.6 0.9 2.7
Reaching for non-moving object 1.4 0.8 2.6
Talking/listening to hand-held device 1.3 0.9 1.8
Drinking from open container 1.0 0.3 3.3
Other personal hygiene 0.7 0.3 1.5
Adjusting radio 0.6 0.1 2.2
Passenger in adjacent seat 0.5 0.4 0.7
Passenger in rear seat 0.4 0.1 1.6
Combing hair 0.4 0.1 2.7
Child in rear seat 0.3 0.04 2.4
Preliminary Results from 100-Car Study
Type of Secondary Task Population Attributable Risk %
Lower CL Upper CL
Reaching for a moving object 1.1 0.97 1.3
Insect in vehicle 0.4 0.3 0.4
Looking at external object 0.9 0.8 1.1
Reading 2.9 2.6 3.1
Applying make-up 1.4 1.2 1.6
Dialing hand-held device 3.6 3.3 3.9
Inserting/retrieving CD 0.2 0.2 0.3
Eating 2.2 1.9 2.5
Reaching for non-moving object 1.2 1.0 1.5
Talking/listening to hand-held device
3.6 3.1 4.1
Drinking from open container 0.04 -0.1 0.2
Odds Ratios for Eyes Off the Forward Roadway (Excluding Mirror Glances)
Total Time EOR
Odds Ratio Lower CL Upper CL
Less than or equal to 0.5 seconds 1.1 0.7 1.9
Greater than 0.5 but less than or equal to 1.0 seconds
1.1 0.8 1.6
Greater than 1.0 but less than or equal to 1.5 seconds
1.1 0.8 1.7
Greater than 1.5 but less than or equal to 2.0 seconds
1.4 1.0 2.0
Greater than 2.0 seconds 2.3 1.8 2.9
OR for EOR 1.6 1.3 1.9
Population Attributable Risk % for Eyes Off the Forward Roadway (Excluding Mirror Glances)
Total Time EOR
Population Attributable Risk %
Lower CL Upper CL
Less than or equal to 0.5 seconds
0.7 0.4 1.1
Greater than 0.5 but less than or equal to 1.0 seconds
1.5 1.0 2.0
Greater than 1.0 but less than or equal to 1.5 seconds
1.6 1.1 2.0
Greater than 1.5 but less than or equal to 2.0 seconds
3.8 3.4 4.3
Greater than 2.0 seconds 18.9 18.3 19.5
OR for EOR 18.3 17.5 19.0
Mean Total EOR Time
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
Crash Near-Crash Incident Baseline
Event Severity
Time Eyes Off Forward Roadway
(Seconds)
Mean # of Eye Glances Away From the Forward Roadway
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
Crash Near Crash Incident Baseline
Severity
Number of Glances Away from the
Forward Roadway
Mean Length of Longest Single Glance
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
Crash Near Crash Incident Baseline
Severity
Mean Length of Longest Glance Away from Forward Roadway
Summary
• The 100 Car approach provides much greater information regarding the pre-crash and crash events than is currently available in crash databases, even those containing detailed crash investigation variables.
• Also, this method provides externally valid data when compared to that obtained on test tracks or in simulators.
Summary
• One of the primary contributions of the 100 Car Study is the creation of an “event” database. This database is similar in classification structure to an epidemiological crash database, but with video, driver, and vehicle data appended. The video and electronic data can be replayed multiple times and at varying frame rates in order to fully understand the nature of each event. The resulting database should be useful for a variety of investigations for the next several years.
Summary
• The 100 Car Study also marks the first time that detailed information on a large number of near-crash events has been collected. Near crashes have two important advantages over crashes. First, they occur much more frequently (e.g., 15 times more often than crashes). Second, every near crash event demonstrates a driver’s successful performance of an evasive maneuver. This may provide additional insight into effective defensive driving techniques and factors, as well as insight into potential countermeasures for these driving situations.
Can large-scale naturalistic data collection assess causation?
• Yes…using epidemiological research methods
• Case-control studies have already been conducted– Use crash and near-crash involvement with baseline
driving epochs as control.– These results are conservative estimate because each
crash/near-crash has multiple baseline epochs to compare» All behaviors per vehicle are matched thus those
drivers who are over-involved in inattention-related C and NC are ALSO over-represented in the baseline database.
• Case-crossover studies, that focus on individual drivers, will be conducted in the future…our next step.
Concluding Remarks
• Epidemiological and empirical research will always be valuable to driving safety research.
• Technology has advanced to a level to give researchers another tool to assess crash causation and develop crash countermeasures in a surface transportation environment.
• This tool will be particularly important for the assessment of the crash risk associated with factors such as driver error, impairment, and distraction.