I.R.I.S.: Real-time Processing of Lidar Data for Road Condition Assessment

Our partner, IMP Bautest AG, provides detailed measurements on the condition of Swiss roads to help decision makers plan maintenance work. After our project, IMP can process the measurement data 150 times faster.

The Swiss road network is 84,000km long with a replacement value of ca. CHF 260 billion, and its maintenance costs public authorities billions of francs every year. Federal, cantonal and local governments need reliable, up-to-date information on the condition of their roads, so that they can plan and budget maintenance work efficiently. Their goal is to minimize both public spending and road closures while maximizing service life and safety.

Many local authorities outsource this road condition survey to companies like our partner, IMP Bautest AG. For strategic life-cycle management, IMP Bautest AG deploys a vehicle that scans the road surface with lidar at the millimeter level. The vehicle, titled I.R.I.S. for Integrated Road Information System, can detect cracks or tell whether rainwater will collect into puddles, for example. The vehicle can travel 80 km/h while maintaining accuracy, so roads won't need to be closed during inspection.

This level of accuracy makes data processing a challenge. IMP’s vehicle carries many different sensors to collect various information, and the company had 8 different computers to handle this data. After each measurement session, it usually took several days to process one day’s worth of collected data, despite parallelisation and powerful hardware. If IMP noticed corrupted data, they’d have to go back and redo the measurements and process data again.

One of our objectives within the project was to simplify and speed up this workflow. Our target was to process the data 120 times faster, which was already an ambitious goal. Yet we didn’t stop there: at the end of the project, our data processing was 150 times faster than before. Instead of several days, it now takes only minutes to obtain the first results. This real-time quality control in the field lets us make adjustments on site and avoid collecting bad or corrupted data.

We started the project by taking a closer look at the data processing steps: we fed in past measurements and reprocessed the data on our own computers. We profiled every step to identify how each microsecond was used, and then we analyzed what could be optimised. We created better algorithms and accelerated them through parallelisation techniques, such as Single Instruction, Multiple Data (SIMD). We also tapped into methods like memory localisation and IO bandwidth optimisation to shave off time wherever we could.

Our collaboration with IMP Bautest AG is a good example of how industry and academia can benefit from each other. Our industry partner gets access to the latest research and data expertise at an affordable cost, while we and our students get a chance to work with real-world problems instead of just publishing theoretical findings. Seeing – and feeling – the results of our work while driving on well-maintained roads is gratifying!

Information

• 

  Simon Felix