Magna | Automotive & Aerospace

Data analysis on automated driving

Magna – Data analysis on automated driving
Know-Center is developing a framework that can be used to combine and analyze measurement data of automated vehicles in cooperation with Magna Steyr‘s development team in Graz. The project "ADDaPT" is funded under FFG‘s "EFREtop“ program line.

Automated vehicles pose a major challenge to the automotive industry. Driver assistance systems such as distance warning, lane departure warning or emergency braking systems must be extensively tested and validated prior to being approved and installed, which is the only way to guarantee absolute reliability.

„It is our goal to create an end-to-end development and manufacturing process of automated vehicles. The efficient aggregation and analysis of measurement data helps us reduce development efforts and costs.“

Dr. Stefan Bernsteiner, Groupleader of ADAS Simulation and Data Analysis at Magna Steyr Fahrzeugtechnik


Lack of synchronicity poses a challenge

The project’s goal is to aggregate various types of data – so-called multimodal data – as automated and efficiently as possible. Video data of camera systems installed inside the vehicle and log data of automotive bus systems were collected for that purpose. Deviation in terms of synchronization and temporal resolution of data to be aggregated states a typical problem. In many cases temporal information isn`t available right away either. During the project, e.g., time stamps for video data had to be extracted from the time reflected on the video stream using machine learning techniques. The video data was then synchronized with signals by the lane change assistant (Blind Spot Information System or “BLIS”).


Vision: A Data Processing Pipeline

Aggregated data was analyzed using Big Data technologies. A so-called Hadoop cluster was used to this end, representing a distributed Big Data framework capable of storing very large amounts of data and processing them at high speed.

Cases in which the system wronlgy indicates overtaking vehicles while in blind spot (false-positive) or does not detect vehicles while in blind spot (false-negative) were identified and that way the accuracy of ‘BLIS’ monitored.

As a next development step, the existing framework will be further extended to allow additional validation, especially such as comfort-oriented driving functions, adaptive cruise control and highway-assist functionality. It’s our vision to build a data processing pipeline which will combine and evaluate data using SQL-like syntax and user-defined regulation.