Start Date
11-12-2016 12:00 AM
Description
The widespread adoption of smart devices and vehicle sensors has the potential for an unprecedented real time assessment of road conditions. The international roughness index (IRI) is an important road profile quality indicator well suited for a crowd based sensing approach. One of the challenges, however, is the heterogeneous nature of sensor measurements from multiple cars that need to be integrated. In this paper, we propose a self-calibration approach that utilizes multiple statistical models trained individually for each car, which in turn get integrated into an overall view of the road segment’s IRI. We evaluate our approach on a dataset collected from seven drives with a total distance of 32 km, with a smartphone equipped car. The dataset contains GPS, accelerometer and gyroscope measurements. Our results show that this approach can reach a mean R² of 0.68 for single car predictions and a R² of 0.75 for combined predictions.
Recommended Citation
Laubis, Kevin; Simko, Viliam; and Schuller, Alexander, "Road Condition Measurement and Assessment: A Crowd Based Sensing Approach" (2016). ICIS 2016 Proceedings. 20.
https://aisel.aisnet.org/icis2016/DataScience/Presentations/20
Road Condition Measurement and Assessment: A Crowd Based Sensing Approach
The widespread adoption of smart devices and vehicle sensors has the potential for an unprecedented real time assessment of road conditions. The international roughness index (IRI) is an important road profile quality indicator well suited for a crowd based sensing approach. One of the challenges, however, is the heterogeneous nature of sensor measurements from multiple cars that need to be integrated. In this paper, we propose a self-calibration approach that utilizes multiple statistical models trained individually for each car, which in turn get integrated into an overall view of the road segment’s IRI. We evaluate our approach on a dataset collected from seven drives with a total distance of 32 km, with a smartphone equipped car. The dataset contains GPS, accelerometer and gyroscope measurements. Our results show that this approach can reach a mean R² of 0.68 for single car predictions and a R² of 0.75 for combined predictions.