Details
| Original language | English |
|---|---|
| Pages (from-to) | 14254-14260 |
| Number of pages | 7 |
| Journal | IFAC-PapersOnLine |
| Volume | 53 |
| Issue number | 2 |
| Publication status | Published - 2020 |
| Event | 21st IFAC World Congress 2020 - Berlin, Germany Duration: 12 Jul 2020 → 17 Jul 2020 |
Abstract
Modern vehicles are equipped with numerous sensors and hence offer an increasing degree of environmental perception. In this work, a method is presented that is able to classify different road types and their conditions based on standard vehicle sensors. Therefore, training and validation data on two routes in urban traffic and on federal highways was gathered using a Volkswagen Golf GTE Plug-In Hybrid. The method uses features based on both frequency and time domain extended with a physical vehicle sub-model. For the classification a decision tree model is trained offline and implemented for online use on target hardware commonly used in modern vehicles. A Bayesian and Markov based filter is used to smooth the output and increase the accuracy of the classification. Since the method is based on sensors that are available in modern vehicles, there is no need for additional hardware, reducing the effort required for implementation. Results show promising classification performance, especially for classifying cobblestone. The three classes of good, medium and bad asphalt labeled relatively precise despite very similar characteristics. Possible applications of the approach could be to adapt vehicles suspension and driving dynamics, to parameterize driver assistance systems, or to update road maps according to their current condition.
Keywords
- Classification, Decision trees, Inertial measurement units, Machine learning, Markov models, Real-time systems, Sensor fusion, Vehicle dynamics
ASJC Scopus subject areas
- Engineering(all)
- Control and Systems Engineering
Cite this
- Standard
- Harvard
- Apa
- Vancouver
- BibTeX
- RIS
In: IFAC-PapersOnLine, Vol. 53, No. 2, 2020, p. 14254-14260.
Research output: Contribution to journal › Conference article › Research › peer review
}
TY - JOUR
T1 - Real-Time Classification of Road Type and Condition in Passenger Vehicles
AU - Beilfuss, Tim
AU - Kortmann, Karl Philipp
AU - Wielitzka, Mark
AU - Hansen, Christian
AU - Ortmaier, Tobias
PY - 2020
Y1 - 2020
N2 - Modern vehicles are equipped with numerous sensors and hence offer an increasing degree of environmental perception. In this work, a method is presented that is able to classify different road types and their conditions based on standard vehicle sensors. Therefore, training and validation data on two routes in urban traffic and on federal highways was gathered using a Volkswagen Golf GTE Plug-In Hybrid. The method uses features based on both frequency and time domain extended with a physical vehicle sub-model. For the classification a decision tree model is trained offline and implemented for online use on target hardware commonly used in modern vehicles. A Bayesian and Markov based filter is used to smooth the output and increase the accuracy of the classification. Since the method is based on sensors that are available in modern vehicles, there is no need for additional hardware, reducing the effort required for implementation. Results show promising classification performance, especially for classifying cobblestone. The three classes of good, medium and bad asphalt labeled relatively precise despite very similar characteristics. Possible applications of the approach could be to adapt vehicles suspension and driving dynamics, to parameterize driver assistance systems, or to update road maps according to their current condition.
AB - Modern vehicles are equipped with numerous sensors and hence offer an increasing degree of environmental perception. In this work, a method is presented that is able to classify different road types and their conditions based on standard vehicle sensors. Therefore, training and validation data on two routes in urban traffic and on federal highways was gathered using a Volkswagen Golf GTE Plug-In Hybrid. The method uses features based on both frequency and time domain extended with a physical vehicle sub-model. For the classification a decision tree model is trained offline and implemented for online use on target hardware commonly used in modern vehicles. A Bayesian and Markov based filter is used to smooth the output and increase the accuracy of the classification. Since the method is based on sensors that are available in modern vehicles, there is no need for additional hardware, reducing the effort required for implementation. Results show promising classification performance, especially for classifying cobblestone. The three classes of good, medium and bad asphalt labeled relatively precise despite very similar characteristics. Possible applications of the approach could be to adapt vehicles suspension and driving dynamics, to parameterize driver assistance systems, or to update road maps according to their current condition.
KW - Classification
KW - Decision trees
KW - Inertial measurement units
KW - Machine learning
KW - Markov models
KW - Real-time systems
KW - Sensor fusion
KW - Vehicle dynamics
UR - http://www.scopus.com/inward/record.url?scp=85105047362&partnerID=8YFLogxK
U2 - 10.1016/j.ifacol.2020.12.1161
DO - 10.1016/j.ifacol.2020.12.1161
M3 - Conference article
AN - SCOPUS:85105047362
VL - 53
SP - 14254
EP - 14260
JO - IFAC-PapersOnLine
JF - IFAC-PapersOnLine
SN - 2405-8963
IS - 2
T2 - 21st IFAC World Congress 2020
Y2 - 12 July 2020 through 17 July 2020
ER -