Road condition monitoring using unsupervised learning based bus trajectory processing

The road infrastructure maintenance is crucial for hassle-free transportation. The proposed work leverages the dense connectivity of public transportation buses for road condition monitoring at a large scale. It uses the buses equipped with GPS and accelerometer as mobile sensors to infer the road surface roughness and the damaged road segments. The vibration features are computed from accelerometer data using the in-bus controller, and trip records are processed offline using a centralized server. The proposal applies the unsupervised learning based Self Organizing Map (SOM) and $k$-means clustering algorithms on the GPS location records and vibration features to infer the road condition. The damaged segments and rough patches of the selected region that requires immediate repairment are suggested. This information can be used to prioritize the repairment based on the available time and budget.

The proposed solution is evaluated using more than 1150 km of trip records collected over four routes of Gujarat state of India. The proposed solution accurately infers the road roughness and identifies the damaged road segments for maintenance. Moreover, the ablation analysis of the proposal is carried out to evaluate the utility of combined execution of SOM and $k$-means algorithms. Further, the feasibility of proposal for a large scale deployment is assessed. The analysis shows that the proposed system is scalable and can process the daily transit data of a metro-city (e.g. 540 buses of the Ahmedabad Municipal Transport Service) using the in-bus controllers and a server.