In this Master of Science thesis we propose and evaluate an algorithm
for estimation of low velocities. Velocity estimation is the same
problem as Doppler-frequency estimation. Most commonly frequency estimation
methods use the fast Fourier transform for estimation. When velocity
estimating is performed on radar measurement of limited amount of
data and low frequency these methods give a poor result.
In this thesis an extended Kalman filter (EKF) with a smoothing window
is evaluated for velocity estimation of low velocities, which also
means low frequencies. The evaluation is made on both computer-generated
measurements and real radar measurements from road vehicles.
The proposed method has on computer-generated signals a mean absolute
error lower than 0.01 meters per second inside the interval from 0.5
meters per second to 4.2 meters per second when the signal to noise
ratio is zero decibel. On real radar measurements the smoothing EKF
works for velocities below 3.5 meters per second.
Our proposed method is compared to the well-known MUSIC algorithm.
On computer-generated measurements the EKF estimator gives a 6dB improvement
on the error compared to the MUSIC algorithm, this performance is
measured at the velocity interval 1.4-2.6 meters per second. One additional
advantage is that the computation time for the EKF smoother is below
three tenths of the MUSIC algorithms.