Even if we are in the 21st century, digital transformation has not fully covered the counties in developing nations – there are still 20 million miles of roads on our planet not under the purview of modern technology. This created a huge problem in the age of technology, where almost everything is digitalized. Nowadays, most people use digital maps to go somewhere new.
So the roads which have not yet been mapped cause many problems for digital maps, particularly for systems developed for autonomous vehicles. The Massachusetts Institute of Technology (MIT) researchers have come up with a solution. They have developed a deep learning method called RoadTracer which will build road maps from aerial images. These maps created from the aerial images are supposed to be 45% more accurate than any existing approaches.
RoadTracer is well-suited to map areas of the world where maps are frequently out of date, which includes both places with lower population and areas where there’s frequent construction. For example, existing maps for remote areas like rural Thailand are missing many roads. RoadTracer could help make them more accurate. – Mohammad Aizadeh, one of the paper’s co-authors told MIT news
They have used NVIDIA Tesla V100 GPUs on the Amazon Web Services cloud and TensorFlow deep learning framework on the neural network on aerial images of 25 cities across six countries in North America and Europe. They have validated their neural work using the same GPUs that they used for the training by mapping 15 other cities. The MIT team said that if the area they are trying to map has buildings, trees, shadows, or anything of that sort, the whole process of deep learning mapping can suffer many errors.
The current mapping efforts, which are done by using aerial images, are often imprecise. What makes RoadTracer different than other digital mapping systems is that the system creates maps step-by-step. The method starts at a known location and uses a neural network to analyze the surrounding area to determine which point is most likely to be the next part of the road. The system then adds that point on the road and repeats the process to trace the way.
Fayven Bastani, an MIT graduate student and co-author of the paper, says, “Rather than making thousands of different decisions at once about whether various pixels represent parts of a road, RoadTracer focuses on the simpler problem of figuring out which direction to follow when starting from a particular spot that we know is a road. This is in many ways a lot closer to how we as humans construct mental models of the world around us.”
They will present their paper in June at the Computer Vision and Pattern Recognition Conference in Salt Lake City, Utah.