Current Non-Contact Road Surface Condition Detection Schemes and Technical Challenges

Yao Ma, Meizhu Wang, Qi Feng, Zhiping He, Mi Tian2022 · DOI

TL;DR

This review paper surveys current non-contact methods for detecting road surface conditions (RSC), crucial for autonomous driving safety. It highlights infrared spectroscopy as a promising technology, while also detailing challenges and future directions for improving RSC detection accuracy, timeliness, and cost-effectiveness.

What Problem Does This Paper Address?

Autonomous driving systems rely heavily on accurate perception of the environment to ensure safety. A critical, yet often overlooked, aspect is the condition of the road surface, specifically its friction coefficient. This coefficient directly impacts a vehicle's braking capabilities, steering stability, and overall maneuverability. Factors like rain, snow, ice, and even road wear can drastically alter friction, leading to dangerous situations such as skidding and increased braking distances. While current autonomous driving sensors (LiDAR, cameras, radar) excel at obstacle detection, they generally lack the ability to reliably assess road surface conditions and predict friction. This gap poses a significant safety risk, necessitating robust and real-time RSC detection methods.

Methods

The paper categorizes and analyzes four primary non-contact RSC detection methods:

  • Infrared Spectroscopy: This method leverages the unique spectral signatures (reflectance and absorption) of different road surface materials and their states (e.g., water, ice, snow) in the infrared spectrum. By analyzing these spectral fingerprints, the system can differentiate between various road conditions.
  • Computer Vision: Utilizing machine learning and deep learning techniques (like Convolutional Neural Networks - CNNs), this approach analyzes images of the road surface to classify conditions. It learns to map visual features to specific RSC states, often by identifying patterns related to wetness, ice, or snow.
  • Optical Polarisation: This technique exploits the way light polarizes upon reflection from different surfaces. Water and ice, being anisotropic, alter the polarization of reflected light differently than dry surfaces, allowing for discrimination.
  • Radar Detection: This method uses radar signals (millimeter-wave, LIDAR, ultrasonic) to probe the road surface. Changes in backscattered signals or the creation of 3D point clouds can reveal information about surface texture, moisture, and ice, which are then correlated with friction coefficients.

The paper also discusses sensor fusion as a means to overcome individual sensor limitations and improve overall performance.

Datasets

The paper references various datasets used in the cited research, including:

  • USGS spectral library: Used for obtaining spectral profiles of typical land features.
  • Custom datasets: Many studies utilize their own collected datasets, often involving images captured from vehicles or test tracks, labeled with ground truth RSC and friction coefficients.
  • MARWIS and RCM411 data: Commercial vehicle-mounted sensors provide data for comparison and validation.
  • RoadSaW dataset: A newly built dataset based on cameras and MARWIS data for estimating friction coefficients.

The paper notes that the availability and quality of datasets are crucial for training robust computer vision models, and highlights the need for more comprehensive datasets, especially for complex road conditions.

Key Insights and Ideas

  • Infrared Spectroscopy's Promise: The paper strongly advocates for infrared spectroscopy as a highly promising technology for RSC detection due to its potential for high accuracy, rapid detection, and ability to identify specific conditions like ice and snow.
  • The "Fingerprint" of Materials: Infrared spectroscopy's effectiveness stems from the fact that different materials and their physical states (water vs. ice vs. snow) have distinct absorption and reflection characteristics across the infrared spectrum, acting like a unique "fingerprint."
  • Computer Vision's Data Dependency: While powerful, computer vision methods are heavily reliant on the quality and quantity of training data. Lighting conditions and camera resolution can significantly impact performance.
  • Polarisation's Simplicity and Limitations: Optical polarisation offers a relatively simple and low-cost approach but often provides qualitative results and requires precise instrument alignment, limiting its practical deployment.
  • Radar's Robustness and Challenges: Radar excels in adverse weather but struggles with precise road surface media identification and can be expensive.
  • The Need for Multi-Sensor Fusion: No single sensor technology is perfect. Combining data from multiple sensors (e.g., cameras, radar, spectroscopy) is essential for robust and comprehensive RSC detection.
  • Environmental and Instrumental Interference: Ambient light, temperature, and instrument parameters (angle, distance) significantly impact detection accuracy, posing major challenges for practical implementation.

Key Results

  • Studies using infrared spectroscopy have achieved accuracies greater than 90% for water, snow, and icing conditions.
  • Computer vision models have demonstrated accuracies exceeding 92% for RSC classification and have shown potential for reducing braking distances by significant percentages (e.g., 25.8% in snow).
  • Polarisation methods have shown accuracies up to 90% in identifying road conditions.
  • Commercial systems like MARWIS and RCM411 are available, offering RSC detection capabilities, though their cost is significantly higher than traditional methods.

Limitations and Future Work

  • Infrared Spectroscopy Challenges: High technical requirements, sensitivity to ambient light and background, and high product prices are noted limitations. Future work should focus on developing more robust algorithms and potentially active light sources to mitigate these issues.
  • Computer Vision Limitations: Performance is highly dependent on classifier training and image quality, making it susceptible to lighting variations and requiring significant computational power. Future work should explore more advanced fusion techniques and robust algorithms.
  • Polarisation Method Constraints: High precision in instrument installation and mechanical complexity limit its large-scale application. Further research is needed to simplify implementation and improve accuracy.
  • Radar Detection Gaps: Difficulty in identifying special RSC (snow/ice) and interpreting complex data are key limitations. Future work should focus on data fusion and advanced interpretation techniques.
  • Data Processing Timeliness: The limited processing power of current in-vehicle devices is a bottleneck. Future work should focus on optimizing algorithms, dimensionality reduction, and leveraging network information.
  • Special Road Conditions: Detecting black ice, turbid water, and layered ice/snow remains challenging. Future research should focus on material classification, thickness measurement, and establishing robust friction coefficient models for these conditions.
  • Cost and Portability: Current advanced RSC detection systems are expensive. Future work should aim to reduce costs through system optimization and potentially leveraging IoT/IoV for data sharing.
  • Correspondence between RSC and Friction: The paper notes a lack of studies directly correlating RSC with friction coefficients, highlighting this as a crucial area for future research to enhance reliability.

Summary updated

BibTeX
@misc{ma2022current,
  author = {Yao Ma and Meizhu Wang and Qi Feng and Zhiping He and Mi Tian},
  title = {{Current Non-Contact Road Surface Condition Detection Schemes and Technical Challenges}},
  year = {2022},
  doi = {10.3390/s22249583},
}
Figure