Monitoring Street Infrastructures with Artificial Intelligence

Exner, J. P., Nalbach, O., & Werth, D.(2020)
In: SHAPING URBAN CHANGE–Livable City Regions for the 21st Century. Proceedings of REAL CORP 2020, 25th International Conference on Urban Development, Regional Planning and Information Society, CORP–Competence Center of Urban and Regional Planning, S. 589-597

Link zur Publikation:


Sensor-based IoT data is enhancing information gathering methods for urban planning in many ways and, due to the growing data pool provided by these sensors, more and more cities and municipalities are consequently putting the use of artificial intelligence-based (AI) methods on their agenda. One area of urban planning that will benefit significantly from the new possibilities enabled by AI is that of infrastructure monitoring. As the topic of the investment backlog of German road infrastructures increasingly pushes into public discourse, many potential areas for application of such a system are opening up. Given the fact that a large part of the German road infrastructure was planned and built several decades ago, and considering that the traffic volume has increased tremendously since then, the urgency in the development of improved maintenance methods is evident: Today’s solutions for infrastructure monitoring are either too labor- intensive, too resource-intensive or too inflexible for the scenario at hand. However, a promising avenue for further research opened up through the advent of mobile communication devices, such as smartphones, in combination with artificial intelligence approaches. This paper describes the methodology applied in the ongoing research project DatEnKoSt, in which these comparatively cheap and sensor-laden devices are used to realize low-cost acquisition methods: Mounting the smartphone in a vehicle, a multi-sensor datastream can be recorded, including, for instance, accelerometer data, GPS coordinates, image or even audio data. From the datastream, features correlated with the road condition can then be extracted, e.g., image processing methods may extract individual cracks from the image data, signal processing can aid analysis of the accelerometer data to determine strength of vibrations, etc.. Using supervised learning methods, these features may be mapped to standardized profiles of the current state of the infrastructure. Even more, predictive methods can, in addition to a mere monitoring of the current state of the infrastructure, enable new ways to provide more precise forecasts and eventually, leveraging optimization algorithms, automatically derive the right maintenance measures for each given situation. The municipal preservation of traffic routes becomes more efficient and sustainable. The methodology enables the potential for further use in the light of real-time as well as predictive road infrastructure monitoring such as winter road services.