RoadPhone

Is our proposed approach feasible?

The previous research team led by Dr Zili has shown promising results using similar approach, i.e. 90% prediction accuracy, more detail see "Smartphone Sensing of Road Surface Condition and Defect Detection". However, in order to keep the input dataset for ML the same size and reduce the uncertainties during ML training, the approach was tested under perfect conditions, for example, the speed of the vehicle remained a constant for all datasets, anomalies are evenly distributed on the route, the tested route was carefully selected, i.e. straight, no turns, completely flat, etc. It is obvious not the case when it comes to the real world, but, at least, it gives the baseline for the vibration-based approach for road pavement inspection and the feasibility of the unsupervised learning in such a scenario.
In the first phase of the RoadPhone project, one of the tasks is to prove the feasibility of the proposed approach in a complete non-controlled variate environment by simply having a glance of collected datasets.

Data collection

Smartphones are installed on the UCC shuttle bus to collect motion data with GPS locations. The buses run on a designated circle route on a 30-minute bases from the edge of the city to the city centre, where the traffic can be very busy. Each run is approximately 6 km long and takes around 16 minutes to complete. The route contains all types of road pavement anomalies that listed in PSCI standard.
During collections, 3-axis time series acceleration data was collected and processed into vibration/location datasets. This step was repeated as many as possible to enrich the unlabelled dataset. On the other hand, a manual inspection was conducted on a representative section, during which pictures were taken for each anomaly identified to keep record. The photos were then classified according to PSCI, calibrated with road anomalies on the GIS system and mark as labelled data for preparation of machine learning training.

Have a glance of the data

Ideally, larger amplitude of the acceleration indicates higher possibility of severe road anomaly. This can be represented by standard derivation of the road section’s acceleration when vehicle runs through the anomaly. As shown in the anomaly map, standard derivations are marked by colour grades, i.e. the darker colour is, the severer the anomaly is. The labelled anomalies and the photos are right on the locations of the dark red.

Vibrations on the route

This section shows another graph that illustrates the large amplitudes of the accelerations are right on the anomalies by drawing accelerations against accumulative distance.