Naturalistic Driving Data Analytics

The 8^th International Workshop on Naturalistic Driving Data Analytics (NDDA)

IEEE Intelligent Vehicles Symposium - IV 2021

July 11, 2021 | Nagoya University, Nagoya, Japan

Join Zoom Meeting

https://zoom.us/j/98035879688?pwd=dUd5b3BYMXBhNFB3WThqNmlEMUMrUT09

Meeting ID: 980 3587 9688

Passcode: 153042

Scope

Understanding driving characteristics, driver behaviors, vehicle performance characteristics, traffic environment and modalities in real world driving context are important for the development of future mobility applications for Intelligent Vehicles. Driver engagement with the vehicle operation, driver-vehicle capabilities for handling demanding traffic situations, traffic management protocols and fuel efficiency are some of the key problems to address using naturalistic driving data.

Naturalistic Driving data collected from various onboard sensors, infrastructure-based sensors, and other emerging data sources provide a wealth of information pertaining to a snapshot of real-world driving context. However, these data streams are inherently heterogeneous due to multimodal nature of sensor suites and data collection platforms used. Therefore, our intention is to investigate intelligent data analytic approaches to produce meaningful inferences from real-world driving data for the safe deployment of intervening technologies in future mobility applications.

Topics

• Data collection

  • • Data needs and specifications for Naturalistic Driving data collection

    • Robust data compression and annotation techniques

    • Naturalistic driving data collection platforms and sharing

• Data analytics

  • • Advanced and automated driving systems data analytics

    • Understanding/ interpretation of naturalistic driver behavior and modeling

    • Collaborative driving algorithms

• Naturalistic Driving applications

  • • Driver-Vehicle Interaction, Driver- ADAS/ADS Interaction

    • Crash risk analysis, reconstruction, modeling, and intervention using naturalistic driving data

    • Integration of driving models in intelligent vehicle system design

    • Assessment protocols for driver health and wellness using naturalistic driving data

    • Fuel efficiency

    • Collaborative, connected, and shared traffic management systems

Objectives

• Understand driver behaviors in various levels of automation from naturalistic driving data.

• Analyze driving conditions and driver-vehicle performance for safety applications and future mobility solutions.

• Explore driving performance metrics for early diagnosis of health issues.

• Requirement analysis for future applications in automated-connected driving, traffic management, and infrastructure design for future mobility.

Invited Talk 1:

Connected Car Data Lakes and Global Deployment

This presentation will discuss how data scientists and developers benefit from cloud services to build right data strategies to accelerate AI/ML development. Breaking data silos is essential to ensure developer collaboration and acceleration of model of development. It is important to have a cohesive and integrated approach to IoT, data management and AI/ML to develop and deploy models to desired edge devices. In this presentation, AWS will cover customer examples using AWS data lake solution that automatically configures the core AWS services necessary to easily tag, search, share, transform, analyze, and govern specific subsets of data across a company or with other external users. Named a leader in Gartner's Cloud AI Developer services' Magic Quadrant, AWS is helping tens of thousands of customers accelerate their machine learning journey. In this talk, AWS will cover how AWS Sagemaker can help reduce the time it takes to aggregate and prepare data for machine learning (ML) from weeks to minutes.

Invited Talk 2:

Borrowing Dagger from Imitation Learning to Efficient Training of Deep Models on Naturalistic Driving Data

This talk will discuss augmented annotation: an iterative, yet fast framework for annotating naturalistic driving data and training deep models. The goal is to enable deep models to slowly learn to imitate expert annotators. The process involves first, auto-labeling the first set of annotation on data using an off-the-shelf deep model. These annotations are then shown to expert human annotators and actions taken by these annotators to change or modify the annotations. Finally, the new model is trained using the corrected annotations. These iterations are repeated till the deep model achieves stable performance. The performance of the proposed framework is evaluated through spatial visualizations of model accuracy after each iteration. The results show significant reductions in the number of false positives - negatives and an increase in the number of true positives after the fourth iteration of the augmented annotation process.

Invited Talk 3:

Using Naturalistic Driving Data to Predict Mild Cognitive Impairment and Dementia: Preliminary Findings from the Longitudinal Research on Aging Drivers (LongROAD) Study

Emerging evidence suggests that atypical changes in driving behaviors may be early signals of mild cognitive impairment (MCI) and dementia. This study aims to assess the utility of naturalistic driving data and machine learning techniques in predicting incident MCI and dementia in older adults. Monthly driving data captured by in-vehicle recording devices for up to 45 months from 2977 participants of the Longitudinal Research on Aging Drivers study were processed to generate 29 variables measuring driving behaviors, space and performance. Incident MCI and dementia cases (n = 64) were ascertained from medical record reviews and annual interviews. Random forests were used to classify the participant MCI/dementia status during the follow-up. The F1 score of random forests in discriminating MCI/dementia status was 29% based on demographic characteristics (age, sex, race/ethnicity and education) only, 66% based on driving variables only, and 88% based on demographic characteristics and driving variables. Feature importance analysis revealed that age was most predictive of MCI and dementia, followed by the percentage of trips traveled within 15 miles of home, race/ethnicity, minutes per trip chain (i.e., length of trips starting and ending at home), minutes per trip, and number of hard braking events with deceleration rates ≥ 0.35 g. If validated, the algorithms developed in this study could provide a novel tool for early detection and management of MCI and dementia in older drivers.

Invited Talk 4:

The Frequency of Lane Change Check Glances With and Without Partial Automation in a Naturalistic Driving Study

Partial automation is becoming more common in production vehicles. Some partial automation features, such as Tesla Autopilot, include automated lane change capabilities. This study examined whether there were differences in the frequency of check glances when drivers made lane changes with and without automation activated during a naturalistic driving study. The results suggest that visual behavior (i.e., visual checks in the direction of lane changes) was similar between manual and automated lane changes. These data, which are the first to investigate the impact of automation on lane change behavior, suggest that drivers might employ similar visual attention strategies during lane changes with and without partial automation.

Organizers