Target tracking and sensor placement for doppler–only measurements


Tezin Türü: Yüksek Lisans

Tezin Yürütüldüğü Kurum: Orta Doğu Teknik Üniversitesi, Mühendislik Fakültesi, Elektrik ve Elektronik Mühendisliği Bölümü, Türkiye

Tezin Onay Tarihi: 2015

Öğrenci: SÜLEYMAN AYAZGÖK

Danışman: UMUT ORGUNER

Özet:

This thesis investigates the problems of target tracking and optimal sensor placement with Doppler-only measurements. First, a single point track initialization algorithm proposed in the literature is investigated for Doppler-only tracking. The initialization algorithm is based on separable least squares method and involves a grid-based optimization. Second, particle filters are considered for Doppler-only tracking and they are compared to an extended Kalman filter (EKF). It is shown that a classical bootstrap particle filter, rather surprisingly, is inferior to the EKF in a Doppler-only tracking scenario. The reasons for this strange behavior are discussed. Then, classical sequential Monte Carlo tools are investigated to improve the behavior of the bootstrap particle filter. In this regard, two new particle filters, namely, a sequential importance resampling particle filter with optimal proposal distribution and a Rao-Blackwellized particle filter are derived and implemented. The results show that, although there are occasional improvements in the particle filter performance for some specific parameter selections, the improvement mechanisms employed are not sufficiently effective to make the particle filters beat EKF. Finally the problem of optimal sensor placement is considered for Doppler-only tracking. A 1D target motion is considered on a road/line segment and the optimization criterion for sensor placement is selected to be the total position Cramer Rao Lower Bound (CRLB) over the road/line segment. The results obtained using numerical optimization tools are utilized to propose a simple sub-optimal sensor placement strategy with explicit formulae for the sensor positions. The proposed strategy is shown to have very close cost values to the optimal strategy.