Photo by Ananya
Many drivers might have grumbled at their navigation system for leading them through pothole-filled roads under the pretext of the fastest route. It is no secret that the conditions of roads in various parts of our country have seen better days. Even as one section of the road is being repaired, heavy traffic rerouted over another section worsens it. Without a reliable system in place for continuously monitoring road conditions, civic authorities are left to play catch-up with road repairs.
In a recent study, researchers from the Indian Institute of Technology Bombay (IIT Bombay) have developed a new application called RoadCare, which uses crowdsourced data to monitor the quality of road surfaces. This application can provide not only early warning about bad road conditions to drivers and suggest the smoothest route to a destination but also help civic authorities in tracking the changes in road quality over time. The study was funded by the Ministry of Human Resources and Development, Government of India, and the Oswal Foundation, Mumbai.
Current techniques employed to survey road conditions are time-consuming and require the use of specialized vehicles that are hard to procure within the limited budget available in developing countries. Instead, RoadCare is a smartphone application that can be used by commuters to share road data while travelling. Based on this data, the app classifies the quality of road sections as ‘good’, ‘medium’ or ‘bad’, depending on their ‘bumpiness’. For example, if a vehicle can smoothly ride over a section of road, even though it has rough patches, the road is classified as ‘good’.
The RoadCare application runs on a type of statistical algorithm called deep neural networks that can learn the relationships between data and categorize it without needing explicit instructions. For RoadCare, the researchers used two types of data from the phones: accelerometer data, which gives information about the movement of the phone, and GPS data, which carries the latitude and longitude coordinates. This data was shared by 20 taxi drivers in Mumbai over six months as they went about their daily trips over a variety of roads.
The researchers fed examples of accelerometer and GPS data for ‘good’, ‘medium’ and ‘bad’ roads from the dataset to the algorithm. The algorithm could then puzzle out the connections between the data and the quality of a road surface. When a user runs the RoadCare app, the algorithm uses the learnt connections to make real-time predictions of the road conditions. Additionally, RoadCare finds the direction of travel from the compass information in GPS data, and provides road quality data to the user only for the road they are on and not the opposite lane.
On testing the model with known road data, the researchers found that it could satisfactorily predict the conditions of the road, misclassifying only about 3% of the good roads. They explain that this misclassification stemmed from those roads that didn’t neatly fall into either the good or bad category.
RoadCare works with data collected from phones held or placed in any orientation. “Since we crowdsource the data, the flexibility of having the phone in any direction is important. So, our algorithms are built to be agnostic to phone orientation,” says Prof Bhaskaran Raman from IIT Bombay, who is one of the researchers involved in the study. Further, the researchers have collected data from a variety of phones, ensuring that the algorithm works well regardless of the phone model.
The predictions made by RoadCare are visualized by marking the roads with coloured lines on Google Maps. Green line indicates ‘good’ roads, violet for ‘medium’ roads and red is for ‘bad’ roads.
Visualization of the road quality for good (in green), medium (in violet) and bad (in red) roads. [Image by Kartavya Kothari (RoadCare team)]
In some cases, the researchers observed that roads in bad conditions were marked as good. This, they say, happened due to the presence of good patches in the road section along with bad, with good count dominating slightly over the count for the bad category. To overcome this problem, they introduced incremental grading on an 11-point rating system — with ‘0’ marking extremely good and ‘10’ marking extremely bad — instead of only three divisions. As the conditions go from good to bad, the colours in the visualization change from dark green to red. This allows for the visualization to convey more nuanced information.
Visualization of the road quality in the 11-point rating system. [Image by Kartavya Kothari (RoadCare team)]
The RoadCare application continuously updates the latest road conditions based on the data provided by the drivers. Hence, it can be used by civic authorities to record the changes in ratings for the road quality over time. They can discern if the quality of any road has deteriorated and direct their resources in ensuring its timely maintenance. So far, RoadCare has used data given by 25 drivers over fifteen months.
Moreover, RoadCare can suggest alternate routes for a smoother ride. “In many scenarios, commuters may want a smoother route, even if longer, over a faster route,” says Prof Raman. “This could be simply for comfort, or due to health reasons where they cannot withstand rough roads for long, or due for commercial reasons such as transporting fragile equipment. In these cases, having road quality information becomes important.”
As next steps, the researchers are working on improving the integration of RoadCare with Google Maps.
“We have recently built a visualization platform which we plan to open to the public, along with the data we collected on Mumbai roads. We also have plans for further improving our algorithms using data from other sensors such as gyroscope,” adds Prof Raman.
This article has been run past the researchers, whose work is covered, to ensure accuracy.