SAFER has been granted research funding for a major investment in developing its large databases with driving data from traffic environments and various driving cases from around the world. The project aims to accelerate the development of automated vehicles and ensure that these are safe once they reach the market. The development will be conducted with the help of machine learning - powerful computers will track different driver states and behavior in the database, instead of a real person doing the work. The result will thus be obtained faster, and probably also with a more precise result.
Knowledge about driver state necessary
Driver impairment, including distraction and inattention, has a major negative impact on traffic safety. Automation is expected to have a positive effect on safety as the driver is taken out of the driving task. However, for first implementations in SAE level 3, the driver is still expected to serve as fall back which requires a fast switch of attention towards the driving task. So, detecting the driver state and developing a better understanding of driver impairment is a key enabler to enhance existing functions as well as to identify new safety relevant factors to be considered in the development of new systems up to automated driving functions.
Machine learning to evaluate 7,5 million km of data
Recent development in machine learning shows potential in recognizing different driver states (e.g. drowsiness), various secondary task engagements (e.g. texting), and driver posture (e.g.out-of-driving position). These algorithms are data-driven and require large amounts of labelled images for training and testing. The SAFER Naturalistic Driving Data platform has been developed over ten years. The datasets cover 7.5 million km of real-world driving in different contexts, countries and vehicle types. In particular, the dataset includes videos collected from cameras pointed to the driver’s face.
The vision is to create a world class vehicle and traffic safety dataset for research and development of active and passive vehicle safety systems. John-Fredrik Grönvall, SAFER’s area manager for Naturalistic driving data tells more:
“The goal of our project is to create an even more valuable database by using the new machine learning technology. When the upgrade is completed, we will be able to use the database to a much higher degree than today, e.g. to validate safety systems in future vehicles that keeps track of the driver’s attention and drowsiness”.
The project will commence shortly and will be placed in SAFER's research area Safety performance evaluation. Autoliv, Chalmers and Smart Eye are included in the project. Vinnova contributes with approximately SEK 1.6 million and the partners contribute with approximately SEK 3.8 million.