IIT Roorkee Researchers are using Computer Vision to Monitor and Improve Railway Tracks
- Researchers from IIT Roorkee are using computer vision technology with drones to monitor the state of railway tracks
- Quite a lot of the computer vision script work has been done in MATLAB
- The results are impressive; the technique works well in varying weather conditions and differing altitudes
Railways are the lifeblood of millions of Indians. You only need to hop on to the Mumbai local to understand how popular and necessary trains are in India. But negligence of infrastructure over the years has unfortunately led to frequent mishaps and accidents.
So researchers from IIT Roorkee have proposed a solution that combines the power of computer vision with drone based technology. Their approach essentially monitors the health of railway tracks to ensure that they meet the compliance standards and more importantly, prevent accidents.
So how does their approach work? As with any machine learning project, it works in several steps:
- Data acquisition: The drone is flown over a pre-defined path and captures images on it’s own at continuous intervals
- Data preprocessing: As you can imagine, these images require a lot of cleaning. Images are resized and cropped in MATLAB, Gaussian smoothing is applied to remove any noise, among other things
- HSV Color Extraction: Since images are captured during different times of the day in varying weather conditions, they are converted into HSV (hue, saturation, value) space for better detection of the railway tracks
- Edge detection: This was used in order to separate the railway tracks from the background
- Track detection: Self explanatory, this step involved using the “bwareaopen” command in MATLAB
- Gauge measurement: This is a critical aspect of any railway track monitoring process. The team generated a bar chart to show the calculated gauge distance (the distance between all the consecutive points of the track)
Below is an image illustrating the technique:
Once the approach was finalized, the researchers then set about testing it on more than thousand images that spanned different weather conditions. The results, shown in the paper (link below), clearly show that this is an effective technique and is not influenced by weather or altitude.
You can read their research paper in full here.
Our take on this
While there has been research in this area using drones previously (also mentioned in the research paper), this novel approach is cost effective and and takes up far less time. The current system being used by the railway authorities is manual, cumbersome and even inconsistent at times. I hope we see a rapid adoption of this technique, which can work alongside the current process to improve and hopefully prevent accidents from happening down the line.
This is a shining example of academia being used to solve a real-life challenge. It serves as an inspiration to all the students and data science enthusiasts that this diverse field has a lot to offer outside of the usual corporate data science work.
Subscribe to AVBytes here to get regular data science, machine learning and AI updates in your inbox!