Vehicle Speed Estimation using Audio Features and Neural Networks

Many car accidents that result in pedestrian deaths or serious injuries are due to their inattention when crossing the street. Pedestrians often get distracted using mobile phones or music players, what prevents them to perceive warning signs and sounds. In this work, we developed a method to estimate the speed of an approaching vehicle using features of the generated acoustic signals. This system can be used as a component of a warning system of potential road risks for pedestrians. We used a single microphone to record audio signals. They were processed to extract features in frequency and time domains that were used as inputs to a neural network. Speed estimation was done using a feed forward neural network. We used several architectures and training algorithms. Results show mean error percentages of 14.57% for speeds from 10 to 40 km/h when using a neural network with two hidden layers.