Compare and contrast the following two articles (2 pages max). Use APA format. (READ 28 pages)

Yubin Xi, Matthew Crisler

A Review of Lane Change Definitions and 1 Identification Methods 2 3 4 5 6 Yubin Xi 7 Graduate Student, Department of Automotive Engineering, Clemson University International Center for 8 Automotive Research, Greenville, SC 29607 USA 9 E-mail: yubinx@clemson.edu 10 Phone: (864) 325-2881 11 12

Matthew Crisler, PhD (Corresponding Author) 13 Research Specialist, Department of Automotive Engineering, Clemson University International Center for 14 Automotive Research, Greenville, SC 29607 USA 15 E-mail: crisler@clemson.edu 16 17 18 19 4 Research Dr. 20 Clemson University International Center for Automotive Research (CU-ICAR) 21 Greenville, SC 29607 22 23

Submitted: November 15, 2012 24 Word Count: 6257 25 Figures and Tables Count: 2*250=500 26 Total Words: 6757 27 28

29

30

31

32

33

34

35

Submitted for presentation at the Transportation Research Board 92th Annual Meeting and inclusion in 36 conference proceedings, Washington D.C., January 2013 37

Yubin Xi, Matthew Crisler 1

ABSTRACT 1 Lane changes are challenging maneuvers and represent an important component of traffic research. 2 Significant efforts have been spent on lane change related research, and various models have been 3 developed to study lane change behavior from different perspectives. In order to identify lane change 4 maneuvers from time series data, researchers have been using different lane change definitions and 5 identification methods, which makes the conclusions drawn from their research dependent on the author’s 6 choice of definition or method. This article reviews lane change definitions and a collection of 7 identification methods, provides a summary of the existing literature and offers information relevant to 8 the selection of a definition or identification method. 9 10

Yubin Xi, Matthew Crisler 2

INTRODUCTION 1 Traffic crash data has shown that the lane change is a challenging driving maneuver and thus has been an 2 important traffic research object (1,2). A large number of studies address lane change related issues from 3 various perspectives. Many of these studies involve efforts to identify and extract lane change segments 4 from time series data such as evaluation of empirical traffic data, development of driver assistance system 5 or reproduction of lane changes for traffic flow models. In order to allow for comparisons across multiple 6 studies, it is important that researchers have consistent methods to define and segment lane changes since 7 the lane change segments captured from the collected data can impact the research outcomes. Two major 8 factors might affect the ability to consistently interpret and compare the results of research involving lane 9 changes. First, authors might use different definitions of lane changes. At the level of individual research 10 projects, the definition is valid for a specific study as long as it enables capturing the desired information 11 about the lane change maneuver. However, varied interpretations of the scope of lane change maneuvers 12 could produce significant variability in results. For example, some studies have addressed the time-course 13 of a lane-change maneuver (3-10); however, the duration of lane change maneuvers in these studies was 14 not consistently defined. This makes comparisons across studies difficult. Establishing a standard lane 15 change definition is beyond the scope of this paper; however, in order to facilitate understanding lane 16 change maneuvers, lane change definitions utilized in the extant literature will be reviewed and 17 characterized. 18

In addition to varied definitions, the methods adopted for lane change identification also vary 19 widely. One common reason for the difference is that many researchers developed the identification 20 methods based on different definitions or without explicitly referring to an existing definition. Another 21 reason is that the availability of input variables used to identify lane changes varies between studies. This 22 is partially due to different data sources. Data were collected from different experimental environments, 23 (traffic simulation, driving simulator, instrumented vehicle or naturalistic traffic recording, etc.) and each 24 environment produces different variable types. For example, drivers’ head and eye movement and 25 steering input data could be available in instrumented vehicles or driving simulators, but are not available 26 when using traffic simulation or naturalistic traffic recording. Vehicle dynamics variables (velocity, 27 acceleration, yaw angle, etc.) are more readily accessible using driving simulators, instrumented vehicles 28 and traffic simulation. In cases when the identification process in one study involves variables that are 29 unavailable or unused in other studies, comparison across different studies becomes difficult. Even if the 30 data were from similar experimental settings, some factors may still influence the availability of data such 31 as the simulation software capabilities or the level of instrumentation. 32

In order to allow researchers to address issues associated with lane change behavior consistently, 33 this article will review the existing literature related to the definition and identification of lane change 34 maneuvers. In this review, a collection of lane change definitions will be described. In addition, several 35 identification methods will be presented, providing information for those who need to identify lane 36 change segments from time-series data. Since the vast majority of papers reviewed here were not 37 explicitly focused on defining or segmenting lane changes, their research objectives and methodologies 38 will be briefly summarized to provide appropriate context. In addition to providing a review of existing 39 lane-change definitions and identification methods, an example application of lane-change identification, 40 driver assessment and training, is described in the context of the current review. 41

LANE CHANGE DEFINITIONS 42 When it comes to research on lane change maneuvers, it is often the case that there is no standard 43 definition that has been consistently adopted by researchers. Therefore when they identify driver’s lane 44 change maneuvers from time-series data, the same lane change could be represented as different segments. 45 This makes research difficult to compare and replicate and further reduces the research credibility (11). 46 Therefore, in order to study lane change maneuvers, it is important to address how lane change maneuvers 47 have been defined. In this section a group of lane change definitions serving different purposes will be 48 described. 49

Yubin Xi, Matthew Crisler 3

It is worth noting that the lane change duration can be divided into the preparation/decision phase 1 and the execution phase. The lane change preparation/decision phase refers to the period of time during 2 which the driver initiates the desire to change lanes and is gathering information on feasibility of 3 changing lanes. Normally there is no noticeable and deliberate steering activity during this phase. The 4 scope of this review is limited to the identification of the lane change execution phase, so the definitions 5 addressing the preparation phase are not included in the review. Worrall and Bullen conducted a study of 6 lane changing behavior at a macroscopic level on multilane highways (4). Lane change pattern (the 7 number of lane changes occurring among all lanes along a given length of road and over a given time 8 span), frequency (the distribution of lane changes between specific lane-lane pairs along a given road 9 length and over a given time span), maneuver length distributions, maneuver time distributions, and gap 10 acceptance behavior were described. Data were collected using 70mm aerial photography taken at 11 different locations. Lane changes were divided into head, lane-change, and tail stages. The head portion 12 refers to the period of time between the moment when the vehicle moves from a straight path and that 13 when it first encroaches the lane line separating the current and the target lanes. The lane change stage 14 follows the head portion and ends when the vehicle body fully crosses the lane line. This is followed by 15 the tail portion which ends when the vehicle resumes a straight path. 16

Chovan et al. addressed the definition of lane change in an analysis of lane-change crashes 17 guiding the development of a crash avoidance system. In the report, lane change refers to a family of 18 maneuvers including simple lane change, merge, exit, pass and weave maneuvers. Lane change was 19 defined as a deliberate and substantial shift in lateral position of a vehicle (5). This definition explicitly 20 excludes unintended drift either within the lane or across lanes. The definition is followed by a model of 21 ideal lane change behavior, partially based on the work of McKnight et al. (12). It comprises (in order of 22 occurrence) checking the legality of the lane change, information gathering and decision making, using 23 signal, and execution of the lane change. One might notice that the model referred is different from the 24 definition used in the same paper since the model includes a lane change decision phase, which might not 25 involve noticeable lateral movement. 26

Winsum et al. studied the relationship between perceptual information and motor response during 27 a lane change in a driving simulator (13). Specifically, it explains the relationship between visual 28 feedback and a driver’s steering actions. Lane changes were defined using a three-phase method. The first 29 phase begins with initiation of steering wheel movement and ends when it is turned to the maximum 30 angle from the neutral position. In the second phase, the steering wheel is turned in the opposite direction. 31 The second phase ends when the steering wheel passes through the neutral position. At this moment the 32 maximum vehicle heading is reached and the vehicle is at its largest deviation from longitudinal direction. 33 In the third phase, the steering wheel keeps turning to reach the second maximum angle (in the opposite 34 direction). This model offers simplicity since steering wheel angle is the only parameter involved in the 35 identification; however, the down side of only using steering wheel angle is that this method might not 36 capture lane changes on curved roads since the steering pattern would be affected by following the 37 contour of the roadway in addition to completing the lane-change maneuver. Also, when the third phase 38 ends, the vehicle has not yet stabilized in the target lane. At the moment the third phase ends, the vehicle 39 heading is on its way back to following the direction of the lane from its maximum deviation. 40

Olsen et al. have closely examined lane changes and provided multiple criteria for lane change 41 initiation and end points (8, 14). In the dissertation, the author addressed three issues: 1. To characterize 42 slow lead vehicle lane change; 2. To develop a predictive model of lane changing; 3. To provide design 43 guidelines for lane change collision warning systems. Driving data were collected using two 44 instrumented vehicles (a sedan and a SUV). The lane change initiation point was identified using one or 45 more of the following four rules: 46

1. Vehicle begins to move laterally relative to the lane; 47 2. Driver initiates a steering input intended to change the direction of the vehicle relative to the 48

lane; 49 3. Driver returns gaze to the forward view after looking in mirrors or looking directly toward the 50

side or rear; 51

Yubin Xi, Matthew Crisler 4

4. Vehicle leaves the lane at least temporarily. 1 In addition, activation of the turn signal is referenced as an auxiliary criterion. According to the 2

author, the turn signal could be used to locate a lane change, but cannot be relied upon as the initiation 3 point since the turn signal activation is not present in all lane changes and does not always represent the 4 initiation point of the maneuver. The completion point, according to the author, was not as critical as the 5 initiation point. However, it affects task completion time. A lateral-velocity-threshold method for 6 completion point identification was suggested by the author, though in practice the end point was 7 determined by data reductionists’ judgements with regard to ‘settling in the lane’. The author did not 8 address whether the velocity threshold method was consistent with reductionists’ judgements. 9

Tijerina et al. studied eye glance behavior using instrumented vehicles(6). In this work, the 10 authors provided an understanding of the drivers’ glance to the road ahead, mirror use, and head rotation 11 during the lane change preparation phase. The study aimed to provide design guidelines for lane change 12 collision avoidance systems. Lane change maneuvers were defined as separate decision and execution 13 phases. The decision phase was defined as the time interval from when the driver desires to change lanes 14 until the initiation of the execution phase by steering input. This duration was used by the driver to gather 15 information for deciding whether or not to change lanes. The execution phase is defined as the interval 16 from the initial steering wheel input until the vehicle is stabilized within the target lane, returns to the 17 original lane, or a crash occurs. 18

Salvucci et al. introduced a real-time system used to predict the occurrence of lane changes (15). 19 The system is able to continuously infer driver’s unobserved lane change intentions from observed 20 behaviors. Data were collected from both a driving simulator and an instrumented vehicle. In this work, 21 Salvucci et al. defined lane change as a segment in which the vehicle starts moving toward another lane 22 and continues, without reversal, through to that lane. By saying ‘without reversal’, the definition 23 emphasizes the completion of the lane change maneuver and excludes aborted maneuvers. In order to 24 differentiate real lane changes from unintended drifts and to define the initiation of lane changes, a 25 minimum threshold of lateral velocity was used. The lane change initiation point is defined as the moment 26 when the vehicle lateral velocity reaches the threshold. The threshold was set to be 0.35m/s which, 27 according to the author, is conservative because a lane change maneuver would take 10s to finish at 0.35 28 m/s (assuming lane width is 3.5m) while the range of mean values of lane change duration is from 3 to 7 29 seconds (14). Although using 0.35 m/s is based on existing observations of lane-change duration, it is 30 clear that the appropriate lateral velocity threshold may vary with driving context since many factors 31 appear to influence lane change duration including road conditions(city road or highway) (6), the 32 presence of a ride-along experimenter (7), and vehicle types (16). 33

Toledo et al. presented a lane change definition to address the influence of the lane change 34 execution phase in the domain of microscopic traffic simulation where lane changes are conventionally 35 modeled as instantaneous events in such an environment (3). This study used naturalistic driving data 36 collected by high-mounted video cameras. Lane change is defined as passing from one lane to the lane 37 immediately next to it. The initiation and completion point are time instances when the subject vehicle 38 begins and ends lateral movement. The authors were also trying to associate lane change durations with 39 various factors including lane change directions, vehicle types and surrounding traffic. 40

When Fitch et al. examined driver’s behavior leading to lane change crashes or near-crashes, lane 41 change was defined as a driving maneuver that moves a vehicle from one lane to another where both lanes 42 have the same direction of travel (17). Data were collected from naturalistic driving using instrumented 43 vehicles. The report did not take the lateral motion onto the shoulder of the road or into an oncoming lane 44 into account. Initiation and completion points are described which were adapted from a study by Lee et al. 45 (8). Three criteria are presented to determine the initiation point of the lane change maneuver. The 46 predominant criterion is when the driver initiates a steering input intended to change the direction of the 47 vehicle relative to the lane. This criterion is supplemented by the second one to accommodate situations 48 when: 1. In-vehicle video is not available; 2. In-vehicle image contrast is low (e.g. night); 3. Lane change 49 occurs on a curved road. The second criterion of initiation point is when the vehicle begins to move 50 laterally relative to the lane. The third criterion takes drivers’ visual search into account. It is when the 51

Yubin Xi, Matthew Crisler 5

driver returns gaze to the forward view after glancing at a rear-view mirror or side window. The lane 1 change completion point is defined simply as the time when the vehicle normalizes in the adjacent lane. 2 In operation, one analyst is involved in determining the initiation and completion point. 3

Table 1 is a summary of the lane change definitions discussed above. As described above, 4 creating a standard definition of lane-change maneuvers is beyond the scope of this investigation. Instead, 5 lane change definitions were classified using the following criteria. 6

1. Explicit initiation and completion points: In general there appear to be two types of lane 7 change definition statements. One is a general statement without explicit information about the 8 duration of the lane change. In other lane change definitions, an initiation point and a 9 completion point are clearly defined. The column ‘Explicit initiation and completion points’ 10 specifies which definitions explicitly define the initiation and end points. 11

2. Data source: Each definition presented is dependent on or related to a specific source of data. 12 Data sources used in the reviewed papers include: driving simulators, instrumented vehicles, 13 traffic simulation and naturalistic traffic recording. The source of data is an important aspect of 14 the definition of a lane change maneuver because certain inputs are only available in specific 15 contexts (e.g. overhead video data will not be available from an instrumented vehicle). 16

3. Required parameters: In order to use the definitions described here to develop an identification 17 method, one must have access to certain information. The parameters that must be available to 18 utilize these definitions effectively are described; however, some authors did not explicitly list 19 which variables were collected and utilized. As such, the variables listed in the following table 20 are derived from the definitions provided in the literature, but may not be entirely consistent 21 with the variables actually used by the original authors. 22

23

Yubin Xi, Matthew Crisler 6

TABLE 1: Characteristics of Lane Change Definitions 1

Explicit Initiation

and Completion

Points

Required Parameters Data source

Worrall and Bullen(4) Yes

heading angle, vehicle dimension, vehicle lateral position, lane

position,

Naturalistic traffic video

John D. Chovan et al.(5) No Vehicle lateral position

Existing data (Crashworthness Data System

and General Estimates System)

W. van Winsum et al.(13) Yes Steering wheeel angle Driving simulator

Olsen et al.(8, 14) Yes

Vehicle position/lateral velocity relative to the lane, steering wheel angle, driver’s vision, directional

signals

Instrumented vehicles

Tijerina et al.(6) Yes

vehicle lane position; steering wheel position, travel speed, turn

signal activation, lateral acceleration; driver eye glance and

head turns.

Instrumented vehicles

Salvucci et al.(15) No Vehicle lateral velocity Driving simulator, instrumented vehicle Tomer Toledo et

al.(3) Yes Lateral velocity/position, Naturalistic traffic video

Fitch et al.(17) Yes Steering angle, vehicle lateral position, driver’s eye glance Instrumented vehicle

2

LANE CHANGE IDENTIFICATION METHODS 3 There are not as many articles that specifically address developing and implementing computational 4 algorithms to automatically identify lane change maneuvers from time-series data. Generally, the 5 identification process was an intermediate step as part of lane change related research. In this section, the 6 objective of each work will be summarized, and important issues such as the implementation process and 7 required parameters will be discussed in order to support decisions regarding the use of the method in 8 specific contexts. The focus of this review is to aid researchers as they make decisions regarding 9 appropriate identification methods for lane change maneuvers. As such, the implementation process and 10 required parameters presented here are intended only to afford an understanding of the method that will 11 aid in determining whether an identification method is appropriate for a given context, and researchers 12 should refer to the original works for further implementation details. It is also worth noting that not all 13 lane change identification efforts involved implementing data processing algorithms to identify lane 14 changes. There are many cases when lane change initiation and completion points were defined 15 subjectively. This can be done by drivers’ or ride-along experimenters’ noting the initiation and 16 completion points of a lane change maneuver or by having data reductionists review the time history data. 17 The subjective methods will also be reviewed at the end of this section. 18

Yubin Xi, Matthew Crisler 7

Bogard and Fancher explained how ACC (Adaptive Cruise Control) influences driving behavior 1 when a lead vehicle changes speed or when a driver decides to change lanes in a report of the FOCAS 2 (Fostering Development, Evaluation and Deployment of Forward Crash Avoidance System) program (18). 3 As part of this program, two lane change identification methods were introduced using GPS data and 4 path-curvature data respectively. The former was briefly discussed and the latter was elaborated and 5 finally used to identify lane changes. 6

GPS data method: Heading angle was one of the five variables recorded by GPS at 2Hz. From the 7 diagram of heading angle vs. time, one can easily see two types of heading angle changes. Smooth 8 changes are due to road curvatures and sharp changes are due to lane changes. However, low sampling 9 frequency and low reliability of GPS recording prevented this from being the primary lane change 10 identification method used. 11

Path-curvature data method: This six-step method can be summarized as: Calculating heading 12 angle and yaw acceleration from path-curvature data; Identifying heading corners (if the absolute yaw 13 acceleration exceeds 0.01 deg/s2 for more than 5s, the mid-points of the zero-crossing time are defined as 14 the heading corners); Fitting a reference line between heading corners and calculating the difference 15 between the heading angle peak and the reference and the area underneath the pulse. If both values exceed 16 defined thresholds, a lane change maneuver is identified. The authors provide a full illustration of how 17 this methodology was applied and how the criteria were developed. The drawbacks are also discussed. 18 This algorithm is based on straight and constant radius road segments. The assumptions upon which the 19 algorithm is based are not fulfilled for many rural roadways. In addition, the algorithm only captures lane 20 changes that occur at velocities above 50 mph, and it will not capture lane changes when a driver enters or 21 leaves a curve during the lane-change maneuver. 22

Based on the hypothesis that a lane change will generate a noisy-sine-wave-like yaw rate signal, 23 Miller and Srinivasan proposed a method to determine a lane change maneuver of heavy trucks based on 24 yaw rate (19). This is one of the few articles which focuses specifically on lane change identification. The 25 method consists of four steps: 1. Bias and noise removal; 2. Sine wave first half cycle determination; 3. 26 Total time span of a lane change determination; 4. Check. Step 1 aims to make the yaw rate sinusoidal 27 signal center around zero and to eliminate ambient noise by setting all data points with a yaw rate of less 28 than 0.05 deg/s to be zero. Step 2 is to examine if the yaw rate signals approximate a sine wave. Step 3 is 29 to find the third zero-crossing point which concludes a complete lane change. Step 4 is to check if the 30 amplitudes of two half cycles are of opposite signs and determine whether the identified period represents 31 a “wandering in the lane”. This algorighm results in four decisions: no lane change, left lane change, right 32 lane change and wander in the lane. According to the author, the model has a detection reliability rate of 33 80% based on 105 video samples. 34

Thiemann et al. proposed a smoothing algorithm for NGSIM trajectory data and investigated 35 lane change dynamics (20). The data were obtained from naturalistic traffic recording. Four situations 36 were filtered out at the beginging of data processing: 1. Lane changes that were too close to each other 37 (using 5s as seperation threshold) ; 2. Lane changes involving on- or off-ramps (only using lane changes 38 on the four left-most lanes); 3. Aborted lane changes; 4. Misjudged lane changes by tracking algorithm. 39 The proposed algorithm addresses the well-defined part of lane changes -- the time span when the vehicle 40 body ‘rides’ on the lane boundary, which is also the lower bound of the lane change duration. One of the 41 most important variables used is the lane index that the vehicle is currently occupying. A certain lane is 42 being used if the mid-point of vehicle front-bumper lies in the lane. If lane index is found to change 43 between two consecutive timepoints, a lane change event can be assumed. Having the vehicle dimension 44 available(width especially), the timepoints when the subject vehicle encroaches the lane line and when it 45 leaves the line were found around the lane change event time. The modal value of lane change duration 46 obtained using this method is approximately 3s and the authors suggest that it might take 5 to 6 seconds if 47 preparation and post-processing phases are included. 48

Knoop et al. analyzed the number of lane changes as a function of the characteristics of the origin 49 and target lane. Their lane change identification method used loop detectors placed on each lane of a 50 three-lane freeway about 100 meters apart (21). Since time, lane index, vehicle speed and vehicle length 51

Yubin Xi, Matthew Crisler 8

were recorded, a vehicle can be re-identified from one detector to the next. Therefore if a vehicle was re-1 identified at a downstream detector on another lane, a lane change was identified. This method is based 2 on the assumption that no driver makes a complete lane change within 100 meters. According to the 3 author, there are two drawbacks associated with this method: firstly this method only works in 4 uncongested traffic conditions (vehicle speed greater than 72 km/h); secondly this method does not give 5 the accurate trajectory. 6

Koziol et al. (22), when trying to evaluate an Intelligent Cruise Control System, proposed a lane 7 change (referred to as ‘Lane Movement’) identification method using degree of curvature data. First, a 8 time window of 8 seconds was used to examine the captured data points at each time step (using a 1 9 second step length). Next, the captured data points were normalized and integrated to find the inflection 10 point. If the point was found, a potential lane change was noted. Then five parameters characterizing a 11 lane change were computed and compared with their boundary values to further identify a lane change. 12 These parameters included: the inflection of the degree of curvature curve; the maximum and minimum 13 values on the degree of curvature curve; the duration between the maximum and minimum degree of 14 curvature; the duration of the entire lane change. A model validation was also performed and yielded an 15 identification rate of 0.78 and a false alarm rate of 0.2. 16

Based on the work of Bogard (18) and Koziol (22), Ayres et al. (23) came up with a vehicle 17 movement identification method to analyze field operational test (FOT) data. This method is able to 18 detect lane changes, turns and curves on different road types using yaw rate and velocity. First, sensor 19 data bias and noise were removed. Then time intervals for potential events (lane changes, turns or curves) 20 were identified using yaw-rate. For each time interval, the heading angle ratio and the lateral position 21 change were calculated and two consecutive time intervals were grouped if their yaw-rate peaks were of 22 opposite signs. Finally the calculated heading angle ratio and the lateral position change were compared 23 with their thresholds to identify lane changes. In addition to the implementation, the authors explained 24 how the thresholds were set, described the algorithm performance and discussed potential ways to 25 improve the algorithm. According to the validation study, the algorithm had an identification rate of 69%. 26 It was able to identify lane changes on a curve but as two separate events. 27

Xuan and Coifman proposed a lane change detection method using vehicles trajectory 28 information obtained from DGPS (Differential Global Positioning System) (24). To begin with, a 29 reference trajectory needs to be established to represent the road geometry. If road geometry information 30 (center line position) is readily available through methods such as GIS (Geographic Information System), 31 one can skip identifying the reference trajectory and find lane change maneuvers by comparing a single 32 trajectory with the existing road geometry. If the source of a reference trajectory is not available, the 33 reference trajectory will be established using the median of all trajectories. First, a curvilinear coordinate 34 system needs to be set up using an arbitrary trajectory captured by DGPS. Then all the other trajectories 35 are resampled and mapped onto the coordinate system. The median of the lateral distance of all 36 trajectories at each point is defined as the reference trajectory. After the reference trajectory is established, 37 two types of lane changes were defined and targeted: mandatory lane change (MLC) and discretionary 38 ane change (DLC). MLCs are found by comparing the reference trajectory with the mean of all candidate 39 trajectories. The mean of lateral positions with respect to the reference trajectory during a single lane 40 change exhibits a sinusoidal pattern This becomes the indicator of the occurrence of a lane change. After 41 correcting for the fact that reference trajectory changes lane, all the trajectories will become relative to 42 the real road and exhibit the normal lane-changing pattern. Then a lateral velocity of 0.3m/s was used as a 43 threshold to identify lane change maneuvers and identify the initiation and completion points of a MLC. 44

In contrast, DLCs are detected based on overtaking maneuvers. Each overtaking maneuver 45 contains two lane change maneuvers. The lane boundary curves were set up as threshold curves which are 46 1.8m from the lane center on both sides. If the vehicle is beyond the thresholds for a certain time and 47 distance period, a candidate overtaking maneuver could be identified. After eliminating the erroneous 48 identifications due to GPS errors, the real overtaking maneuvers are found.Then the same lateral-velocity 49 technique used to identify MLCs is applied to find DLCs from overtaking maneuvers. It is also stated that 50 the lateral-velocity criterion is subject to change during congestion. 51

Yubin Xi, Matthew Crisler 9

As mentioned above, lane change identification work can also be done using subjective methods 1 based on researchers’ needs. In Salvucci et al.’s study of driver’s control and eye movement during lane 2 changes(9), a semantic method was used to identify a lane change maneuver. In a driving simulator, a 3 multi-lane highway environment was simulated and participants were instructed to report the intentions 4 and completions of lane changes. In cases where a participant failed to report, an experimenter would 5 define the lane change based on when the initiation and completion points seemed apparent. A similar 6 work was done by Hanowski when studying driver fatigue using instrumented vehicles(25). As an 7 auxiliary method to identify critical incidents in a database, the driver was instructed to use an incident 8 pushbutton after the incident had just occured and then had data analysts review the time period around 9 the incident location. In both methods drivers were aware of data collection process, but the authors 10 suggest that these methods were effective. These methods are presented as a reminder that it may be 11 appropriate to manually identify lane changes from recorded data. 12

Table 2 summarizes the characteristics of the lane change identification methods described above. 13 The required parameters are listed because successful application of lane change identification method is 14 dependent on the fact that all required variables are easily and accurately available. For subjective 15 methods, required parameters refer to the subjective action needed to identify lane changes. 16 17

TABLE 2 Lane Change Identification Methods Characterization 18 Required Parameters Data source

Objective Methods

Bogard and Fancher(18) Method using GPS: Heading

angle; Method using path- curvature: Path-curvature, velocity

Instrumented vehicle

Miller and Srinivasan (19) Yaw rate Instrumented vehicle

Thiemann(20) Vehicle dimension; lane index; vehicle position Naturalistic traffic recording

Knoop(21) Vehicle passing time; lane index; vehicle speed; vehicle length Naturalistic traffic recording

Koziol(22) Degree of curvature Naturalistic traffic recording Ayres(23) Yaw rate; Velocity Naturalistic traffic recording

Xuan and Coifman(24) Vehicle lateral position Instrumented vehicle Subjective Methods

Salvucci(9) Verbal protocol data/experimenter’s judgement Driving simulator

Hanowski(25) Driver’s activation Instrumented vehicle 19

DISCUSSION AND CONCLUSION 20 This literature review could serve as both a inventory of relevant efforts and a selection guide for those 21 who need to define and identify lane change maneuvers. This work also represents a starting point for 22 developing a standard definition of lane change maneuvers and a set of methods to enable researchers to 23 implement consistent methods for lane change identification. 24

It can be seen from the lane change definition section that the driver’s steering input and the 25 vehicle lateral movement are frequently used to define the lane change maneuvers (execution phase). In 26 addition to these triggers, the vehicle heading angle is also used. In this sense, the lane change execution 27 phases in different definitions can be mapped to similar durations. However, the lane change preparation 28 phase has not been as rigorously defined. Since the lane change definition is the foundation of the lane 29 change identification, a good lane change definition should use variables that can be easily and accurately 30 obtained. Lane changes can be challenging maneuvers, therefore they may be relevant indicators of driver 31

Yubin Xi, Matthew Crisler 10

performance; however, in order to address performance metrics associated with lane change maneuvers, it 1 is important to use an appropriate definition of the lane change maneuver that is capable of identifying the 2 time-course of the maneuver. The review presented here could be utilized to guide researchers in this and 3 other contexts with respect to utilizing an appropriate lane change definition and identification method. A 4 specific example application is to use lane change maneuvers to assess driver’s performance in a clinical 5 setting. This context has unique requirements for lane change related analyses including: 1. Data are 6 usually obtained from driving simulators or instrumented vehicles; 2. Expensive head position or eye 7 tracking devices are not common in such an environment. Therefore, participants’ head/eye position data 8 are assumed to be unavailable; 3. With similar performance, the identification algorithm should be as easy 9 to implement as possible to reduce the workload of data reductionists; 4. Identification algorithm does not 10 necessarily have to capture the lane change preparation phase, but should capture durations between 11 vehicle’s first lateral movement and its stabilization in the target lane. By filtering lane change 12 identification models using the above criteria, Miller and Srinivasan’s method (19) might be selected for 13 use in the context of driver assessment in the clinical environment. 14

Even though a significant research effort has been devoted to defining and identifying lane 15 changes, there is still significant work to be completed before providing an integrated lane change 16 definition and identification method that will allow for consistency across research studies. Future works 17 include: 1. establishing a systematic definition of lane changes. The definition should explicitly define the 18 duration of both lane change preparation phase and execution phase. It should also take into account of 19 following features: a. lane change direction (right lane change or left lane change); b. mandatory lane 20 changes and discretionary lane changes; c. handling of multiple-lane changes (crossing multiple lanes at 21 one time); d. handling of curved roadway segments; 2. developing a corresponding lane change 22 identification method. The identification method should be able to identify the phases and lane-change 23 components described in the definition. Additionally, the required parameters should be easily and 24 accurately obtainable in the contexts for which the lane change definition is applicable. 25 26 REFERENCES 27 1. Wang, Jing-Shiarn, and Knipling R. R. Lane Change/Merge Crashes Problem Size Assessment and 28

Statistical Description. Publication DTNH22-91-C-03121. U.S. Department of Transportation, 1994 29 2. Sen, B., Smith J. D., and Najm W. G. Analysis of lane change crashes. Publication DOT-VNTSC-30

NHTSA-02-03. U.S. Department of Transportation, 2003 31 3. Toledo, T., and Zohar D. Modeling duration of lane changes. Transportation Research Record: 32

Journal of the Transportation Research Board, No. 1999, Transportation Research Board of the 33 National Academies, Washington, D.C., 2007, pp. 71-78. 34

4. Worrall, R., and Bullen A. An empirical analysis of lane changing on multilane highways. Highway 35 Research Record, Vol., No. 303, 1970, pp. 30-43. 36

5. Chovan, J. D., Tijerina L., Alexander G., and Hendricks D. L. Examination of lane change crashes 37 and potential IVHS countermeasures. Publication DOT-VNTSC-NHTSA-93-2, U.S. Department of 38 Transportation, 1994 39

6. Tijerina, L., Garrott W. R., Stoltzfus D., and Parmer E. Eye glance behavior of van and passenger car 40 drivers during lane change decision phase. Transportation Research Record: Journal of the 41 Transportation Research Board, No. 1937, Transportation Research Board of the National 42 Academies, Washington, D.C., 2005, pp. 37-43. 43

7. Hetrick, S. Examination of driver lane change behavior and the potential effectiveness of warning 44 onset rules for lane change or “side” crash avoidance systems: Virginia Polytechnic; 1997. 45

8. Lee, S. E., Olsen E. C. B., and Wierwille W. W. A comprehensive examination of naturalistic lane 46 changes. Publication DTNH22-00-C-07007, U.S. Department of Transportation, 2004 47

9. Salvucci, D. D., and Liu A. The time course of a lane change: Driver control and eye-movement 48 behavior. Transportation Research Part F: Traffic Psychology and Behaviour, Vol. 5, No. 2, 2002, pp. 49 123-132. 50

Yubin Xi, Matthew Crisler 11

10. Olsen, E. C. B., Lee S. E., Wierwille W. W., and Goodman M. J. Analysis of distribution, frequency, 1 and duration of naturalistic lane changes. Proceedings of the Human Factors and Ergonomics Society 2 Annual Meeting, 2002, pp. 1789-1793. 3

11. Savino, M. R. Standardized names and definitions for driving performance measures: Tufts 4 University; 2009. 5

12. McKnight, A. J., and Adams B. B. Driver education task analysis. Volume I: task descriptions. Final 6 report. 1970. 7

13. Van Winsum, W., De Waard D., and Brookhuis K. Lane change manoeuvres and safety margins. 8 Transportation Research Part F: Traffic Psychology and Behaviour, Vol. 2, No. 3, 1999, pp. 139-149. 9

14. Olsen, E. C. B. Modeling slow lead vehicle lane changing: Virginia Polytechnic Institute and State 10 University; 2003. 11

15. Salvucci, D. D., Mandalia H. M., Kuge N., and Yamamura T. Lane-change detection using a 12 computational driver model. Human Factors: The Journal of the Human Factors and Ergonomics 13 Society, Vol. 49, No. 3, 2007, pp. 532-542. 14

16. Hanowski, R. J., Wierwille W. W., Garness S. A., Dingus T. A., Knipling R. R., and Carroll R. J. A 15 Field Evaluation of Safety Issues in Local/short Haul Trucking. Proceedings of the Human Factors 16 and Ergonomics Society Annual Meeting, Vol. 44, No. 20, 2000, pp. 3-365-363-368. 17

17. G.M.Fitch, S.E.Lee, S.Klauer, J.Hankey, J.Sudweeks, and T.Dingus. Analysis of Lane Change 18 Crashes and Near-Crashes. Publication DTNH22-00-C-07007 U.S. Department of Transportation, 19 2009 20

18. Bogard, S., and Fancher P. Analysis of data on speed-change and lane-change behavior in manual 21 and ACC driving. Publication DTNH22-94-Y-47016 U.S. Department of Transportation, 1999 22

19. Miller, R. J., and Srinivasan G. Determination of Lane Change Maneuvers Using Naturalistic Driving 23 Data. 19th International Technical Conference on the Enhanced Safety of Vehicles, 2005. 24

20. Thiemann, C., Treiber M., and Kesting A. Estimating acceleration and lane-changing dynamics from 25 next generation simulation trajectory data. Transportation Research Record: Journal of the 26 Transportation Research Board, No. 2088, Transportation Research Board of the National 27 Academies, Washington, D.C., 2008, pp. 90-101. 28

21. Knoop, V. L., Hoogendoorn S., Buisson C., and Shiomi Y. Quantifying the number of lane changes in 29 traffic: Empirical analysis. Transportation Research Board 91st Annual Meeting; 2012; Washington 30 DC; 2012. p. 18. 31

22. Koziol, J., Inman V., Carter M., Hitz J., Najm W., Chen S., Lam A., Penic M., Jensen M., Baker M., 32 Robinson M., and Goodspeed C. Evaluation of the Intelligent Cruise Control System Volume II – 33 Appendices. Publication, 1999 34

23. Ayres, G., Wilson B., and Le-Blanc J. Method for identifying vehicle movements for analysis of field 35 operational test data. Transportation Research Record: Journal of the Transportation Research Board, 36 Vol. 1886, No. -1, 2004, pp. 92-100. 37

24. Xuan, Y., and Coifman B. Lane change maneuver detection from probe vehicle DGPS data. IEEE 38 Intelligent Transportation Systems Conference; Toronto, Ontario, Canada; 2006. pp. 624-629. 39

25. Hanowski, R. J. The impact of local/short haul operations on driver fatigue. Blacksburg, VA: 40 Virginia Polytechnic Institute and State University, 2000. 41