Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Lightweight Regression Model with Prediction Interval Estimation for Computer Vision-based Winter Road Surface Condition Monitoring

Risto Ojala, Alvari Seppänen

TL;DR

Winter road safety requires estimating tyre-road friction from camera imagery. SIWNet delivers a lightweight CNN that predicts a friction factor $\hat{f}$ and a prediction interval via a dedicated head that outputs $\hat{\sigma}$, trained with a truncated-normal likelihood. It achieves competitive point-estimate accuracy with substantially fewer parameters and FLOPs than larger CNNs, while providing calibrated uncertainty estimates on the SeeingThroughFog dataset. The work also releases open-source code, highlighting practical on-board deployment for robust vehicle control in winter conditions.

Abstract

Winter conditions pose several challenges for automated driving applications. A key challenge during winter is accurate assessment of road surface condition, as its impact on friction is a critical parameter for safely and reliably controlling a vehicle. This paper proposes a deep learning regression model, SIWNet, capable of estimating road surface friction properties from camera images. SIWNet extends state of the art by including an uncertainty estimation mechanism in the architecture. This is achieved by including an additional head in the network, which estimates a prediction interval. The prediction interval head is trained with a maximum likelihood loss function. The model was trained and tested with the SeeingThroughFog dataset, which features corresponding road friction sensor readings and images from an instrumented vehicle. Acquired results highlight the functionality of the prediction interval estimation of SIWNet, while the network also achieved similar point estimate accuracy as the previous state of the art. Furthermore, the SIWNet architecture is several times more lightweight than the previously applied state-of-the-art model, resulting in more practical and efficient deployment.

Lightweight Regression Model with Prediction Interval Estimation for Computer Vision-based Winter Road Surface Condition Monitoring

TL;DR

Winter road safety requires estimating tyre-road friction from camera imagery. SIWNet delivers a lightweight CNN that predicts a friction factor and a prediction interval via a dedicated head that outputs , trained with a truncated-normal likelihood. It achieves competitive point-estimate accuracy with substantially fewer parameters and FLOPs than larger CNNs, while providing calibrated uncertainty estimates on the SeeingThroughFog dataset. The work also releases open-source code, highlighting practical on-board deployment for robust vehicle control in winter conditions.

Abstract

Winter conditions pose several challenges for automated driving applications. A key challenge during winter is accurate assessment of road surface condition, as its impact on friction is a critical parameter for safely and reliably controlling a vehicle. This paper proposes a deep learning regression model, SIWNet, capable of estimating road surface friction properties from camera images. SIWNet extends state of the art by including an uncertainty estimation mechanism in the architecture. This is achieved by including an additional head in the network, which estimates a prediction interval. The prediction interval head is trained with a maximum likelihood loss function. The model was trained and tested with the SeeingThroughFog dataset, which features corresponding road friction sensor readings and images from an instrumented vehicle. Acquired results highlight the functionality of the prediction interval estimation of SIWNet, while the network also achieved similar point estimate accuracy as the previous state of the art. Furthermore, the SIWNet architecture is several times more lightweight than the previously applied state-of-the-art model, resulting in more practical and efficient deployment.
Paper Structure (18 sections, 5 equations, 7 figures, 6 tables)

This paper contains 18 sections, 5 equations, 7 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Road surface condition monitoring
  4. Prediction intervals for deep learning regression models
  5. Research gap
  6. Methods
  7. Problem formulation and dataset
  8. Model architecture
  9. Loss function for prediction interval head
  10. Training and testing
  11. Results
  12. Model size and computational load
  13. Point estimate head performance
  14. Prediction interval head performance
  15. Ablation study
  16. ...and 3 more sections

Figures (7)

  • Figure 1: Illustration of the sensor installations, with the camera at the windshield and the road friction sensor at the bumper.
  • Figure 2: Transformation of the area in front to bird's-eye-view.
  • Figure 3: Number of samples per date in the dataset.
  • Figure 4: Friction factor values in the dataset, with a zoomed view providing a clearer depiction of the under-represented values.
  • Figure 5: Samples from the utilised data, featuring images, summary of road surface state, and corresponding ground truth friction factor values.
  • ...and 2 more figures