01732655

Component

http://www.trb.org/Publications/Blurbs/179567.aspx

Unsealed roads remain the backbone of the economy for many countries around the world. Controlling dust on these roads requires useful and cost-effective monitoring. As existing surveying methods are expensive and the condition of unsealed roads can change rapidly, there is a need for cost-effective options to survey these roads. This paper introduces a low-cost qualitative methodology to estimate dust emissions from unsealed roads using automated analysis of photographs taken from camera-equipped smartphones. Visual indicators extracted from photographs can be analyzed using machine-learning techniques to classify dust severity. Image analysis cannot provide information on dust particles directly since the particles’ dimensions are much smaller than a single image pixel. However, dust leaves a fingerprint in the image that acts as a proxy for the combined size and density of the particles. Indicators of road dust in an image include attenuation of high spatial frequencies, reduced saturation, and reduced color gamut. These effects are caused by the scattering and diffraction of light by dust. Challenges to accurate classification include image blur, large image regions containing only light dust, and regions of sky and road within the image. Image preprocessing, classification techniques that assess dust intensity within dusty regions and texture analysis will mitigate these issues. The research concluded that the image analysis methods are showing strong potential for their intended purpose. It demonstrated that a number of techniques could be utilized for the image analyses, each having strong points and limitations within specific operating environments.

01730920

English

pp 439-444

2019-9

Transportation Research Circular

12th International Conference on Low-Volume Roads

Digital/other

Figures; Photos; References

Dust; Image analysis; Monitoring; Photographs; Unpaved roads

Highways; Pavements

TRIS, TRB, ATRI

Mar 3 2020 3:17PM