Christian Claudel's presentation from the 2017 CTR Symposium http://ctr.utexas.edu/ctr-symp/

1. IMU-based road condition
monitoring
Christian Claudel
Assistant Professor, CAEE
University of Texas, Austin

2. Motivations
 Maintaining a transportation network is one of the
mandates of a transportation agency
 Monitoring is usually done with expensive equipment
(profilers)
 Tradeoff between accuracy and refresh rate

3. Inertial measurement Units
 Proposed solution: using Inertial Measurement Units
(IMUs) onboard vehicles for monitoring road condition
 IMUs are or will be part of future CVs (BSM requirement)
 Could enable all CVs to generate road condition data

4. Challenges
1. IMUs generate vector measurements (acceleration, rotation
vector, magnetic field), relative to coordinates of the device.
We need self-calibrating IMUs.
2. Road condition indices, for example the Present
Serviceability Rating (PSR) are usually obtained by
humans. How can we reliably estimate the road condition
indices from vertical acceleration timeseries data?

5. System components (Vehicular system)

6.  To determine the vertical acceleration component, we need to
compute the orientation of the IMU within the vehicle
 Achieved by computing a rotation matrix mapping the
coordinates of the device to the coordinates of the vehicle. This
computation is done automatically, after a few minutes driving.
Automatic Calibration of IMUs
(Mousa, et al. ACM/IEEE IPSN‘16)
csR

7. In device frame In vehicle frame
 Automatic calibration allows
IMUs to be quickly and easily
retrofitted to existing vehicles for
CV applications
 Applications include road
condition monitoring, but also
traffic monitoring

8. Road condition monitoring principle
• We intend to use Supervised Learning to automatically
learn the road condition grading process used by
humans (when determining the PSR)
Quantities
derived from
acceleration data
Road condition
metrics (PSR)

9. Timeline
• Identify possible regressor variables:
- Average acceleration power spectral density in different
frequency bands
- Number of high acceleration events (potholes) per unit
distance
- Peak to peak acceleration level
- Vehicle speed
• Collect measurement data on a subset of the road
network for which PSRs are available
• Train ML framework to estimate PSRs from regressors

10. Research challenges
• Are regressors relevant to explain the variations of PSR
data? How should the IMU be configured, and are
additional measurements needed (ex: microphone)? Is
the signal to noise ratio of conventional IMUs sufficient?
• How fast does the ML framework converge/how much
data is needed for training?
• Is the ML computed framework transposable to different
vehicles (within the same category)?

11. Preliminary data (UT)

12. Preliminary data (SwRI)
Smooth Road

13. Preliminary data (SwRI)
Rough Road

14. Vision
• The road condition monitoring system could be part of a
larger monitoring system, including traffic monitoring
• Could be integrated with Bluetooth/WiFi readers, or
Roadside Equipments (RSEs). The readers would
receive measurement data from almost all vehicles
vs