National Academies Press: OpenBook
« Previous: Chapter 4 - Develop a Methodology for Analyzing Data
Page 42
Suggested Citation:"Chapter 5 - General Guidelines for Video Data Analysis." National Academies of Sciences, Engineering, and Medicine. 2011. Feasibility of Using In-Vehicle Video Data to Explore How to Modify Driver Behavior That Causes Nonrecurring Congestion. Washington, DC: The National Academies Press. doi: 10.17226/14509.
×
Page 42
Page 43
Suggested Citation:"Chapter 5 - General Guidelines for Video Data Analysis." National Academies of Sciences, Engineering, and Medicine. 2011. Feasibility of Using In-Vehicle Video Data to Explore How to Modify Driver Behavior That Causes Nonrecurring Congestion. Washington, DC: The National Academies Press. doi: 10.17226/14509.
×
Page 43
Page 44
Suggested Citation:"Chapter 5 - General Guidelines for Video Data Analysis." National Academies of Sciences, Engineering, and Medicine. 2011. Feasibility of Using In-Vehicle Video Data to Explore How to Modify Driver Behavior That Causes Nonrecurring Congestion. Washington, DC: The National Academies Press. doi: 10.17226/14509.
×
Page 44
Page 45
Suggested Citation:"Chapter 5 - General Guidelines for Video Data Analysis." National Academies of Sciences, Engineering, and Medicine. 2011. Feasibility of Using In-Vehicle Video Data to Explore How to Modify Driver Behavior That Causes Nonrecurring Congestion. Washington, DC: The National Academies Press. doi: 10.17226/14509.
×
Page 45

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

C H A P T E R 5 General Guidelines for Video Data AnalysisBecause all the data sets discussed in this report used video cameras to record driver behavior, certain rules need to be followed. All the naturalistic data sets identified in this proj- ect are protected by the VT IRB, a committee that approves, monitors, and reviews research involving humans with the aim of protecting the rights and welfare of the study partici- pants. These approvals have been granted to the team mem- bers. In video-based data sets, IRB protocols are restrictive regarding who may access the data because of the inherent personal information (e.g., faces of drivers or location of res- idences). Access is often limited to key personnel associated with the original data collection project, making it difficult for an outside entity to gain access to an institution’s data. It is straightforward for the VTTI team to access and use in-house data, as well as outside data that have IRB approval, but special issues need to be addressed when conducting research using video data because human beings are involved. In case any institute performs a similar task using raw video data and other raw kinematic vehicle data, the following pro- cedures are necessary. Assuming the IRB protocol permits access to a contractor, the contractor must be capable of complying with IRB require- ments. At a minimum, IRB approval from the entity main- taining the data set is required; approval from the contractor’s IRB may also be necessary. In some instances, the original IRB agreement will not permit sharing of any personally identi- fiable data. In these cases, the contractor must perform the analyses without the personally identifiable data (e.g., with- out video). Once IRB approval is granted, the institution per- forming analysis on video-based data sets must possess the capability to work efficiently with data sets from outside sources. The contracting institution must have a secure com- puting facility dedicated to the storage, retrieval, processing, and analysis of data. Permissions must be established so that the videos and any associated data files for the project may be accessible only to researchers and other personnel involved in the project.42Each data set to be analyzed for this project typically has two data types associated with it. When applicable, video files will exist in a certain format, such as MPEG2, Audio Video Interleaved (AVI), or MP4. Additionally, each video file will be associated with a parametric data file, such as a comma- separated values (CSV) file, SQL database, or binary data file. This parametric data file contains information collected from the vehicle sensors, such as velocity, acceleration, time of day, and radar data. The contracting institution must have the capability to access and manipulate the various data formats provided. For researchers to determine congestion factors, several tools are required to view and examine a video-based data set. The institution performing the analysis must have access to and expertise with the tools necessary to perform manipula- tion of parametric data. Commercially available tools, such as MatLab and SAS, are commonly used to transform data, identify events of interest, and perform statistical analyses of large data sets. Once events of interest are identified (e.g., crashes and near crashes), the data reduction process begins. Data reduction is the process of manually viewing video and parametric data to validate events and derive additional measures. This process requires an institution that has facili- ties and tools for viewing synchronized video and parametric data, as well as a tool that links parametric data to video data and provides an interface for entering additional data into the data set. Because no tool designed specifically for this purpose is commercially available, the contracting institution will need to have the capability to develop a custom solution. Some research groups (e.g., VTTI) have developed software packages for data reduction that could easily be modified to meet the needs of this project. Finally, a team of data reductionists will need to be trained on proper treatment of video data involving human subjects and the software used to reduce and analyze data. Reduc- tionists should be educated on the goals of the project, the nature of the events, and the protocol for coding variables in

43a robust manner. They work independently to interpret events and record additional variables into the database. Examples of variables obtained through data reduction include weather, traffic conditions, and roadway type. Input from reduc- tionists should be monitored and supervised by a reduction manager, and measures of inter-rater and intra-rater relia- bility should be used to ensure that quality control is main- tained. The contracting institution should have experience with data reduction, as well as facilities within which data reduction can be performed. The reduced data provide a rich source of information to relate driver behavior to surrounding roadway, weather, and traffic conditions. Driver status information (e.g., secondary tasks, driver distraction, driver apparent impairment, and driving habits) is available from the video data. Vehicle status information (e.g., vehicle speed, acceleration and decelera- tion, and lane orientation) is available as part of the in-vehicle kinematic data. A proper linkage of driver and vehicle vari- ables to external variables (traffic, weather, work zone, and accident data) through location and time information will enable an in-depth study of the relationship between driver behavior and travel time reliability. General Guidelines for Video Data Reduction According to the general guidelines discussed earlier, the most challenging effort related to data reduction involves video data. Having conducted multiple naturalistic driving studies involving video data, VTTI has developed a mature procedure to organize, reduce, and analyze large-scale natu- ralistic driving data. According to different research objec- tives, the threshold values selected may vary from one data reduction to another, but the basic principle is consistent and the procedure is uniform. The step-by-step description of video data reduction in each project has been detailed indi- vidually in Chapter 4. Following is a summary of the general guidelines for video data reduction. There are three primary steps in video data reduction. The first is to scan the raw data to identify potential safety events. VTTI developed its own software (DART) that reads and cal- culates numerical data to detect if events of interest have occurred. Therefore, the first step in data reduction is to run the event trigger program. The variables examined include velocity, acceleration (longitudinal or lateral), swerve, TTC (calculated from range and range rate between the subject vehicle and other vehicles detected by radar), and yaw rate. The software reads through the raw numeric data and extracts an epoch of data when it detects a parameter exceeding the trigger threshold set by researchers in advance. Different lengths of the epoch and threshold values can be adopted according to different attributes of the subject vehicle (e.g.,heavy commercial trucks or light cars) and study goals. Details of the trigger threshold values for VTTI studies are listed in Tables 4.3, 4.6, and 4.9 in Chapter 4. The second step is to check the validity of the potential event epochs generated in step 1 through visual inspection. As described in Figure 4.7 (1), invalid events and valid events (including conflicts and nonconflicts) are differentiated. Invalid events are false alarms in which sensor readings are spurious because of a transient spike or other anomaly. Valid events are events in which recorded dynamic-motion values occur and are verifiable in the video and other sensor data from the event. Valid events are further parsed as conflicts and nonconflicts. Conflicts are occasions in which a crash, near crash, incident, or conflict happens. Nonconflicts are occasions in which the trigger values satisfy the thresholds but the driver makes a maneuver and is in complete control of the vehicle. Examples can be a hard brake without any obvious reasons or a high swerve value from a lane change. In most cases, nonconflicts reflect aggressive driving habits or styles. In step 2, valid events are labeled for the next step. The final step is to apply the data dictionary to the vali- dated events. In this step, a data dictionary created by the researcher in advance is used to extract useful information from the raw data. Data reductionists code results while they watch the video by choosing correct answers from pull-down menus for each research question in the data dictionary. Vari- ables included in the data dictionary should reflect the inter- ests of investigators and the research objectives. This is the step in which the identification of driver behavior that affects congestion is extracted. Reductionists can plot real-time charts of any selected variable in the DART—such as range, velocity, and acceleration—simultaneously with the play of the associated video clips. When studying the relationship between driver behavior and nonrecurring congestion, the LOSs and traffic density before and after the event, as well as the contributing factors, are important. The LOS or traffic density before and after can be used in a simulation envi- ronment or in statistical modeling (details of travel time modeling are included in the previously submitted Task 5 Report) so that the events that generate congestion can be iden- tified. Contributing factors, especially those that are driver- and vehicle-related, are judged by reductionists as correctable and avoidable or otherwise. For example, data reductionists decide if a crash or near crash was caused by one of the fol- lowing reasons: (1) the driver was conducting a secondary task, such as talking on a cell phone, dining, adjusting the radio, or talking to passengers; (2) the driver was inattentive to the road, either distracted by something inside or outside the vehicle or simply daydreaming; (3) the driver was driving under the influence (such as drugs, alcohol, or medicine); (4) the driver made a wrong decision; or (5) the driver has questionable driving habits or driving skills. Simultaneously, other possible

44Figure 5.1. Quality assurance and quality control flowchart at VTTI.

45contributing factors, such as weather, traffic flow, traffic den- sity, traffic control, road profile, surface condition, and light- ing condition, are identified. If available from the video clip, the status of other involved vehicles can also be examined and coded. A similar procedure is applied to reduce baseline epochs. The baseline epochs were used as a benchmark in which driver behavior can be compared with that in a safety-related event. According to the design of the researchers, a certain number of baseline epochs of selected length are decided. For exam- ple, the baseline epochs in Project 7 were 60 s long with a total of 1,072 epochs, whereas in Project 8 the baseline epoch was 30 s long and 456 were selected. To ensure high-quality data reduction, a quality control procedure needs to be established. On the one hand, differ- ences between data reductionists should be minimized. On the other hand, data reductionists should ensure that their judgments are consistent over time. Accordingly, inter-rater and intra-rater reliability are tested regularly. Before a data reduction effort starts, several test events are selected and coded by expert analysts (e.g., the principal investigator [PI] of the project or a data reduction director who has extensive experience with data reduction). Next, each reductionist is asked to code the same events. Their coding results are com- pared with those of the experts, and discrepancies are noted and discussed. The validation results help to determine (1) if reductionists are correctly coding the events; (2) if certain reductionists have systematic bias on certain questions; and (3) if the data dictionary is rationally composed and easily understood. Associated training, supervision, and modifica- tion of the data dictionary are applied if needed. For the intra- rater reliability test, one example event is selected for each category representing crash, near crash, proximity, and crash- relevant events. Data reductionists are required to code these events to the dictionary at the beginning of the process. They code the same event again after a period of time (varying from a week, a month, or a year, depending on the full length of the data reduction effort). The results are compared with the original reduction results. If there are differences, reduc- tionists are notified to review their work and make necessary adjustments. In summary, three-step data reduction is effective in pro- cessing large-scale video data and other numeric data. Theselection of threshold values is always a compromise between exhaustiveness and false events. Lower trigger values will capture the maximum number of potential events, but the trade-off is a higher chance of false-positive events, nonconflict events, and less severe conflicts. Similarly, a higher trigger value will result in a higher percentage of valid events but will gener- ate some omissions. A careful examination of threshold values by experts is highly recommended before data reduction starts. For studies such as Project 2 and Project 5, which were aimed at testing onboard alarm systems, data reduction is rel- atively more straightforward. When an alert was triggered or the driver comment button was pushed, the main system logged the triggered summary file, and the video system was notified and captured a retrospective clip of video data with transition counts, histograms, errors, and other trip summary information recorded to a trip-summary log. Because data were already organized by trips and alarms during collection, the scanning step to identify potential events from numerical data is unnecessary. Data reductionists can start data reduc- tion from viewing epochs of video data of events and then coding variables to the data dictionary. Data associated with the reduced data include demographic information (driver’s age-group), lighting condition, road type, and traffic density. Additionally, numerous secondary behaviors (behaviors besides driving) are coded and include such actions as con- versing, adjusting the radio, speaking on a cell phone, and brushing one’s hair. The same modeling method for estimat- ing travel time reliability and judging if driver behavior is cor- rectable can be applied. Figure 5.1 summarizes a typical data reduction process (2).References 1. Hanowski, R. J., M. Blanco, A. Nakata, J. S. Hickman, W. A. Schaudt, M. C. Fumero, R. L. Olson, J. Jermeland, M. Greening, G. T. Hol- brook, R. R. Knipling, and P. Madison. The Drowsy Driver Warning System Field Operational Test, Data Collection: Final Report. Report DOT HS 811 035. NHTSA and Virginia Tech Transportation Insti- tute, Blacksburg, Va., 2005. 2. Lerner, N., J. Jenness, J. Singer, S. G. Klauer, S. Lee, M. Donath, M. Manser, and N. Ward. An Exploration of Vehicle-Based Monitoring of Novice Teen Drivers: Draft Report. Virginia Tech Transportation Institute, Blacksburg, Va., 2008.

Next: Chapter 6 - Measuring Travel Time Reliability »
Feasibility of Using In-Vehicle Video Data to Explore How to Modify Driver Behavior That Causes Nonrecurring Congestion Get This Book
×
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

TRB’s second Strategic Highway Research Program (SHRP 2) Report S2-L10-RR-1: Feasibility of Using In-Vehicle Video Data to Explore How to Modify Driver Behavior That Causes Nonrecurring Congestion presents findings on the feasibility of using existing in-vehicle data sets, collected in naturalistic driving settings, to make inferences about the relationship between observed driver behavior and nonrecurring congestion.

The report, a product of the SHRP 2 Reliability focus area, includes guidance on the protocols and procedures for conducting video data reduction analysis.

In addition, the report includes technical guidance on the features, technologies, and complementary data sets that researchers can consider when designing future instrumented in-vehicle data collection studies.

The report also highlights a new modeling approach for travel time reliability performance measurement across a variety of traffic congestion conditions.

An e-book version of this report is available for purchase at Google, Amazon, and iTunes.

  1. ×

    Welcome to OpenBook!

    You're looking at OpenBook, NAP.edu's online reading room since 1999. Based on feedback from you, our users, we've made some improvements that make it easier than ever to read thousands of publications on our website.

    Do you want to take a quick tour of the OpenBook's features?

    No Thanks Take a Tour »
  2. ×

    Show this book's table of contents, where you can jump to any chapter by name.

    « Back Next »
  3. ×

    ...or use these buttons to go back to the previous chapter or skip to the next one.

    « Back Next »
  4. ×

    Jump up to the previous page or down to the next one. Also, you can type in a page number and press Enter to go directly to that page in the book.

    « Back Next »
  5. ×

    To search the entire text of this book, type in your search term here and press Enter.

    « Back Next »
  6. ×

    Share a link to this book page on your preferred social network or via email.

    « Back Next »
  7. ×

    View our suggested citation for this chapter.

    « Back Next »
  8. ×

    Ready to take your reading offline? Click here to buy this book in print or download it as a free PDF, if available.

    « Back Next »
Stay Connected!