This content is not included in your SAE MOBILUS subscription, or you are not logged in.

Automatic Maneuver Boundary Detection System for Naturalistic Driving Massive Corpora

Journal Article
2014-01-0272
ISSN: 1946-4614, e-ISSN: 1946-4622
Published April 01, 2014 by SAE International in United States
Automatic Maneuver Boundary Detection System for Naturalistic Driving Massive Corpora
Sector:
Citation: Sathyanarayana, A., Sadjadi, S., and Hansen, J., "Automatic Maneuver Boundary Detection System for Naturalistic Driving Massive Corpora," SAE Int. J. Passeng. Cars – Electron. Electr. Syst. 7(1):149-156, 2014, https://doi.org/10.4271/2014-01-0272.
Language: English

Abstract:

Towards developing of advanced driver specific active/passive safety systems it is important to be able to continuously evaluate driving performance variations. These variations are best captured when evaluated against similar driving patterns or maneuvers. Hence, accurate maneuver recognition in the preliminary stage is vital for the evaluation of driving performance. Rather than using simulated or fixed test track data, it is important to collect and analyze on-road real-traffic naturalistic driving data to account for all possible driving variations in different maneuvers. Towards this, massive free style naturalistic driving data corpora are being collected. Human transcription of these massive corpora is not only a tedious task, but also subjective and hence prone to errors/inconsistencies which can be due to multiple transcribers as well as lack of enough training/instructions. These human transcription errors can potentially hinder the development of algorithms for advanced safety systems, and lead to performance degradations. In order to prevent these errors from propagating, an automatic maneuver boundary (also activity) detection system (or in short a MAD tool) utilizing filterbank analysis of vehicle dynamic signals is proposed. Using a minimal set of generic vehicle dynamic sensor information, it is shown that the MAD tool can match human transcription to an accuracy of up to 99%. The tool performs equally well on CAN-bus signals as well as inertial sensor information from a portable device. This simple, accurate, and computationally efficient tool can help mitigate human transcription errors and make valuable data from large naturalistic driving corpora more accessible.