cs.CVMar 30, 2026

Intelligent Road Condition Monitoring using 3D In-Air SONAR Sensing

Authors: Amber Cassimon, Robin Kerstens, Walter Daems, Jan Steckel

Organizations: All authors are with the Department of Electronics and ICT Engineering Technology, Cosys-Lab Research Group, University of Antwerp, Antwerp, Belgium Flanders Make Strategic Research Centre, Lommel, Belgium

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

In this paper, we investigate the capabilities of in-air 3D SONAR sensors for the monitoring of road surface conditions. Concretely, we consider two applications: Road material classification and Road damage detection and classification. While such tasks can be performed with other sensor modalities, such as camera sensors and LiDAR sensors, these sensor modalities tend to fail in harsh sensing conditions, such as heavy rain, smoke or fog. By using a sensing modality that is robust to such interference, we enable the creation of opportunistic sensing applications, where vehicles performing other tasks (garbage collection, mail delivery, etc.) can also be used to monitor the condition of the road. For these tasks, we use a single dataset, in which different types of damages are annotated, with labels including the material of the road surface. In the material classification task, we differentiate between three different road materials: Asphalt, Concrete and Element roads. In the damage detection and classification task, we determine if there is damage, and what type of damage (independent of material type), without localizing the damage. We are succesful in determining the road surface type from SONAR sensor data, with F1 scores approaching 90% on the test set, but find that for the detection of damages performace lags, with F1 score around 75%. From this, we conclude that SONAR sensing is a promising modality to include in opportunistic sensing-based pavement management systems, but that further research is needed to reach the desired accuracy.

Figures & tables

Fig. 1: An image showing the mounting of the eRTIS SONAR sensor (bottom left of the sensorbox) on a vehicle. The sensorbox is mounted on the back of the vehicle, looking behind.
Fig. 3: The size of each dataset, for each fold in the dataset.
Fig. 4: The distribution among different damages in the dataset. Note the logarithmic X-axis.
Fig. 5: The distribution among different materials in the dataset. Note the logarithmic X-axis.
Fig. 6: The vector preprocessing pipeline (top, blue) and the image preprocessing pipeline (below, green). Both pipelines are fed using time-domain, PDM-encoded SONAR recordings for all channels, and output either a feature vector or energyscape, depending on the pipeline.
Fig. 7: A schematic representation of the used neural network architecture. The stem convolution is marked in green in the top-left, the schematic shows two residual modules marked in red, and one downsampling module with a blue background in the top-right. Finally, the global average pooling and linear layers are included at the bottom-right.
ModelTest κ\kappaValidation κ\kappaTraining κ\kappaTest F1Validation F1Training F1
Logistic Regression40.34%±1.22%40.34\%\pm 1.22\%48.00%±3.18%48.00\%\pm 3.18\%45.18%±0.68%45.18\%\pm 0.68\%71.77%±0.90%71.77\%\pm 0.90\%75.73%±1.73%75.73\%\pm 1.73\%74.66%±0.77%74.66\%\pm 0.77\%
Decision Tree76.82%±1.90%76.82\%\pm 1.90\%92.03%±1.23%92.03\%\pm 1.23\%89.49%±0.48%89.49\%\pm 0.48\%89.03%±0.63%89.03\%\pm 0.63\%95.93%±0.47%95.93\%\pm 0.47\%94.88%±0.16%94.88\%\pm 0.16\%
Random Forest76.84%±0.69%76.84\%\pm 0.69\%93.25%±1.46%93.25\%\pm 1.46\%91.16%±0.42%91.16\%\pm 0.42\%89.44%±0.26%89.44\%\pm 0.26\%96.64%±0.54%96.64\%\pm 0.54\%95.82%±0.19%95.82\%\pm 0.19\%
Support Vector Machine74.67%±1.66%74.67\%\pm 1.66\%89.97%±2.59%89.97\%\pm 2.59\%88.03%±2.02%88.03\%\pm 2.02\%88.18%±1.71%88.18\%\pm 1.71\%95.02%±1.73%95.02\%\pm 1.73\%94.28%±1.74%94.28\%\pm 1.74\%
Multi-Layer Perceptron77.52%±2.25%77.52\%\pm 2.25\%91.95%±2.21%91.95\%\pm 2.21\%89.42%±1.34%89.42\%\pm 1.34\%89.12%±1.18%89.12\%\pm 1.18\%95.95%±1.08%95.95\%\pm 1.08\%95.03%±0.76%95.03\%\pm 0.76\%
Gradient-Boosted Trees78.10%±1.40%\mathbf{78.10\%\pm 1.40\%}93.58%±1.83%\mathbf{93.58\%\pm 1.83\%}91.37%±1.07%\mathbf{91.37\%\pm 1.07\%}89.55%±0.60%\mathbf{89.55\%\pm 0.60\%}96.79%±1.01%\mathbf{96.79\%\pm 1.01\%}95.95%±0.51%\mathbf{95.95\%\pm 0.51\%}
TABLE I: The results of the different models on the material classification problem, averaged across random initalizations and folds. All numbers are given as μ±σ\mu\pm\sigma . The best result in each column is marked in bold.
ModelTest κ\kappaValidation κ\kappaTraining κ\kappaTest F1Validation F1Training F1
Logistic Regression25.88%±1.03%25.88\%\pm 1.03\%30.14%±2.35%30.14\%\pm 2.35\%27.88%±1.25%27.88\%\pm 1.25\%49.16%±0.58%49.16\%\pm 0.58\%52.69%±1.17%52.69\%\pm 1.17\%51.20%±0.31%51.20\%\pm 0.31\%
Decision Tree54.98%±1.51%54.98\%\pm 1.51\%65.23%±2.36%65.23\%\pm 2.36\%61.34%±0.86%61.34\%\pm 0.86\%68.42%±0.78%68.42\%\pm 0.78\%75.76%±1.60%75.76\%\pm 1.60\%73.18%±0.31%73.18\%\pm 0.31\%
Random Forest57.87%±1.46%57.87\%\pm 1.46\%65.96%±2.03%65.96\%\pm 2.03\%62.11%±1.01%62.11\%\pm 1.01\%70.69%±0.69%70.69\%\pm 0.69\%76.40%±1.39%76.40\%\pm 1.39\%73.85%±0.49%73.85\%\pm 0.49\%
Support Vector Machine52.64%±1.25%52.64\%\pm 1.25\%59.90%±1.63%59.90\%\pm 1.63\%56.56%±0.81%56.56\%\pm 0.81\%68.33%±1.01%68.33\%\pm 1.01\%73.15%±1.40%73.15\%\pm 1.40\%70.94%±0.91%70.94\%\pm 0.91\%
Multi-Layer Perceptron54.15%±1.82%54.15\%\pm 1.82\%63.54%±2.15%63.54\%\pm 2.15\%59.44%±1.41%59.44\%\pm 1.41\%69.25%±0.95%69.25\%\pm 0.95\%75.83%±1.25%75.83\%\pm 1.25\%73.22%±0.62%73.22\%\pm 0.62\%
Gradient-Boosted Trees58.00%±2.38%\mathbf{58.00\%\pm 2.38\%}66.66%±2.66%\mathbf{66.66\%\pm 2.66\%}62.52%±2.10%\mathbf{62.52\%\pm 2.10\%}71.74%±1.15%\mathbf{71.74\%\pm 1.15\%}77.75%±1.54%\mathbf{77.75\%\pm 1.54\%}75.06%±0.90%\mathbf{75.06\%\pm 0.90\%}
TABLE II: The results of the different models on the damage detection problem, averaged across random initalizations and folds. All numbers are given as μ±σ\mu\pm\sigma . The best result in each column is marked in bold.
Fig. 8: A confusion matrix of the test set for the gradient boosted trees classifier, trained on the first fold of the dataset for material classification, using a random seed of 0.
Fig. 9: A confusion matrix of the test set for the gradient boosted trees classifier, trained on the first fold of the dataset for damage classification, using a random seed of 0.
More in this topic

Structural Health Monitoring

Structural Health Monitoring uses sensor data and computational models to detect, locate, and assess damage in infrastructure and engineered systems.

See all 80 papers

Explore similar work

CardsList

  1. Radiometric fingerprinting of object surfaces using mobile laser scanning and semantic 3D road space models

    Mar 11, 2026Benedikt Schwab, Thomas H. KolbeRemote Sensing3D Scene Representation

    Chair of Geoinformatics, TUM School of Engineering and Design, Technical University of Munich, Arcisstraße 21, Munich, 80333, Germany

  2. HSI-Road Relabeled: Surface-Aware Road-Scene Segmentation

    Sep 10, 2026Imad Ali Shah, Imran Mehmood, Enda Ward +3Remote Sensing Image SegmentationSemantic Segmentation

    School of Engineering and Ryan Institute, University of Galway, Ireland · Valeo Vision Systems, Tuam, Ireland

  3. Pixel-Level Pavement Distress Assessment Using Instance Segmentation

    May 25, 2026Logan Dewick, Bibesh Pyakurel, Kong Pheng Yang +2Image SegmentationInstance Segmentation

    University of Wisconsin - Green Bay · Computer Science Department, University of Wisconsin–Green Bay, Green Bay, WI, USA