Detecting bad weather conditions is included in one of the challenges that are faced in building self driving cars. According to the report released by US Department of Transportation, wet pavements have caused 959,760 crashes and the death of 4,789 people from 2002 and 2012. This figure sums up to 74% of all weather-related crashes in the US and makes up to 23% of all vehicle crashes in the country.
Fortunately, researchers in IEEE (Institute of Electrical and Electronics Engineers) have figured out how to detect a slippery road by analyzing audio feedback from the car’s tires and recurrent neural networks (RNN), an artificial intelligent network of computers. The experiment was done by attaching a shotgun microphone at the rear tire of 2014 Mercedes CLA. The car travelled at different speeds around the Greater Boston area in Massachusetts.
The initial results from the tests show precisions with unweighted average recall (UAR) of 93.2% across all vehicle speeds. The microphone is also able to receive audio feedback from the vehicles that were passing by beside it.
It is the first time that an artificial intelligence is used to detect road conditions. However, it is not the first time that somebody tried to detect road conditions.
Technical University of Madrid in 2014 has done experiments by using support vector machine (SVM) to analyze the sounds when the tyre meet the road and classify the different sound made by the asphalt. However, as there were limited range of surface types and unrelated audio input like the sound of pebbles bouncing against the tyres could lead to false predictions.
University of Toyama in 2014 tried to use surveillance cameras on cars to look for any road reflections caused by other cars’ headlights, however, this way requires other cars to be present on the same road at the same time. On the other hand, this method is working perfectly in detecting road conditions in fog, snow and poor road lighting conditions.