Slobodan Dukanović presents Neural-network-based acoustic vehicle speed estimation

On 2020-11-06 13:00:00 at The Seminar will be held Online (see below)
We address vehicle speed estimation based on the sound that vehicles produce
while passing by the microphone. The sound offers numerous advantages with
respect to the vision (microphones are less expensive, consume less energy and
require less storage space than cameras, they are not affected by visual
occlusions and lighting conditions, they are easier to install and maintain,
and have low wear and tear).
Our speed estimation approach is based on regression of speed-dependent
features. We consider two features: pseudo-distance and attenuation profile.
Pseudo-distance has a V-shape, with speed-dependent slope, centred at the
pass-by instant of the vehicle. Attenuation profile is related to a gradual
reduction in the intensity of the sound signal as a function of the
vehicle-to-microphone distance. The speed-dependent features are predicted via
a two-stage (coarse-fine) regression, both realised using neural networks.
estimation is performed using the samples of the predicted features. The method
is trained and tested on a dataset collected for this research, with 239 sound
files of eight different vehicles passing by the microphone with constant speed
(range from 30 to 105 km/h) preset by the cruise control system. The root mean
square error of speed estimation is about 10.5 km/h (pseudo-distance) and 8.4
km/h (attenuation profile). Experiments show that the accuracy of speed
estimation significantly decreases in windy conditions.

The seminar will be via Zoom:
Meeting ID: 512 449 9156
Passcode: ezj7X9
Za obsah zodpovídá: Petr Pošík