Promoting excellence in mobility engineering

  1. FISITA Store
  2. Technical Papers

Development of the Driver Distraction Monitoring System based on In-vehicle Sensor Data
FISITA2016/F2016-ACVB-001

Authors

Yang, Seokyoul*, Lee, Cheolha*, Im, Seongsu*, Kim, Jinhak*, You, Byungyong*

* Hyundai Motor Company, South Korea

Abstract

Research and/or Engineering Questions/Objective

The driving patterns for all of drivers are different from each other, even though they are running on same road. If a driver distraction monitoring system based on driving pattern is detected using an absolute criterion, it could not get reliability for drivers using the system because of the incomprehensible false alarm or error detection. The objective of this study was to propose the detection system learning driver’s driving pattern from in-vehicle sensor data and providing different decision criterion.

Methodology

First, the system acquires driving and surrounding environment data from the sensors basically equipped in vehicle as well as ADAS (Advanced Driver Assistance System) sensors and then extracts their features. In pre-learning step, the states of lane keeping and lane change are distinguished from the lane information using a camera sensor, and driving modes are classified according to the relation with a preceding vehicle and shape of road (straight / curve). The learning step for each driver was comprised of three types, which are the relation between the preceding and an ego vehicle, the driver’s steering wheel control, and relation between the lane information and an ego vehicle. Obtaining driver’s probabilistic model based on the three types of data, and that is compared with the driving model from real-time driving data in order to periodically monitor the driver’s current state (normal / distraction).

Results

To evaluate the system performance of the proposed detection algorithm, setting the scenario driving with the secondary tasks and without tasks and then, evaluation tests were implemented in the real on-road environment as well as the simulator. As a result, the system performance was effectively verified through ROC (Receiver Operating Characteristic) curve, and it was found out the similar performance in the real on-road and the simulation environment.

Limitations of this study

This study drew a conclusion from the limited number of drivers and their data patterns. More generalized decision criterion is required to evaluate more various drivers’ current state based on the real-time driving pattern and learned driving pattern previously.

What does the paper offer that is new in the field including in comparison to other work by the authors?

This study was to monitor driver’s state without additional sensor equipment only for this system and we proposed the system using a machine learning to detect each driver’s distraction having different driving pattern for the various drivers.

Conclusions

This study implemented the system monitoring the driver’s current state based on probabilistic model from in-vehicle sensor data and then the system performance was verified in the simulator and on real-road and also having validity of all of two environments.

Key Words driver distraction; support vector machine (SVM); in-vehicle sensor; secondary task Research

Add to basket