‘Human error is the most important cause of road accidents, but the impact of road condition and maintenance [on road accidents] is not negligible’ as a recent study from the European Union reports. Key causes are, amongst others, ‘localized anomalies, such as ruts, potholes and depressions, whose unpredictability make them dangerous for drivers’ (Source: EU Road Surface. Economic and Safety Impact of the lack of regular road maintenance, 2014).
According to the EU Road Surfacereport, maintenance expenditures are huge: For instance, the European Union spent about 20.000 million EUR for road maintenance in 2011. In Austria, 350 million EUR were spent for maintaining Austrian motorways by the responsible company, ASFINAG. These huge expenditures make road damage detection a commercially-relevant field for technology development to support road maintainers and local authorities in better localizing road damage.
The aim of the Automotive Demonstrator V1 – Broken Road Indicatoris to provide insights into road conditions by exploiting vehicle sensor- and map data. The main data source is vehicle movement data.
This vehicle movement data is acquired by using a data logging device developed at VIRTUAL VEHICLE. It is based on a BeagleBone Black single plate computer which is connected to the vehicle’s on-board diagnostics interface. Additionally, the data logger is equipped with acceleration, rotation and GPS sensors, all of them built in a especially designed, custom cape. The data logger can easily be installed in the vehicle by the drivers themselves, without the need of technical assistance. The device turns on automatically at the vehicle’s start and starts recording data. Likewise, it shuts down and turns off automatically if the vehicle’s engine is turned off.
For the automotive demonstrator, data was generated by multiple drivers on routes located mainly in the greater Area of Graz. This data was uploaded to the AEGIS big data platform for event detection which happens as follows:
In a first step the raw data files of the individual sensors of each vehicle are merged and all trips contained in the data are extracted. In our case, a “trip” is defined as the data collected between engine start and engine stop. All extracted trips are resampled to a fixed, regular-spaced time grid of 10Hz and written to a separate dataset. Next, the coordinate system of the sensors is aligned with the coordinate system of the vehicle for each of these trips and the data is written into another dataset. This is especially relevant as the position of the data logger in respect to the vehicle is unknown and additionally can change between trips.
Finally, all trips prepared in this way will be loaded to infer broken road events. These events are stored for each individual trip in a new dataset. For the detection of such road damage events, an artificial event-signal using the z-acceleration and gyroscope (pitch) is computed. This event-signal has large values in presence of a speedbump/pothole and low values in other situations. Hence, speedbumps and potholes are detected as periods in measurement data where these values are high.
Many of the steps in the previous data pipeline create enriched, intermediate results which may also be of interest and value for other projects hosted on the AEGIS big data platform. Consequently, we store such results in their own datasets, thereby enabling access and use by other projects to facilitate the establishment of a “data market”.
The most relevant and valuable dataset generated by this demonstrator is however the dataset containing broken road information. The data therein can be visualized directly on the AEGIS platform using either the “marker” or the “heatmap” visualization type (Figure 1).
To compute the heatmap, we use a kernel density estimator that computes the local density of “broken road events”. The resulting density is additionally normalized by amount of data available for each region. Or computations are implemented “online” – meaning that we can add new data without the need to recompute everything, allowing for fast updates of the visualization if new data arrives.
The visualization itself is embedded into a dashboard – both visualization types allow navigation and zoom-in and out. They are of great value to all people interested in gaining information about areas of road damage. Especially, our work supports employees responsible for road maintenance in their decision-making process.
Figure 1: Marker- and heatmap visualization of road damage based on vehicle data
Blog post authors: VIF