Researches have been done to detect dangerous road conditions. According to US Department of Transportation, wet roads have caused 959,760 crashes and caused 4,789 deaths in over 10 years period of between 2002 and 2012. This figure sums up to 74% of all weather related crashes in US and makes up 23% of all vehicle crashes in the country.
Researchers from IEEE (Institute of Electrical and Electronics Engineers) used recurrent neural networks (RNN), a type of artificially intelligent network of computers, to detect how slippery a road is. On top of that, they also attach a shotgun microphone to analyze the audio feedback from the car’s tires. In this process, they used 2014 Mercedes CLA driven during different weather road condition and speed around the Greater Boston area in Massachusetts.
Researchers from Technical University of Madrid in 2014 used support vector machines (SVM), a type of machine learning model, to analyze the sounds made from the tyre meeting the road and classify them accordingly. However, they found that the range of surface types were limited and unrelated audio input such as the pebbles sound bouncing against the tyres could lead to false predictions.
Researchers from University of Toyama in Japan have conducted the same study in October 2012. They were experimenting with a system that used surveillance cameras on cars by looking at the road reflections from the headlights of other drivers’ passing by. It showed an accurate road conditions in fog, snow and poor light, however, it required other drivers to be on the road too for this method to work.