ON ESTIMATION OF NON-STATIONARY MOTION PARAMETERS IN VIDEO SEQUENCES

摘要

Motion estimation of nonstationary parameters in video sequences is considered. It is based on the time-frequency analysis and the subspace-based line detection (SLIDE) algorithm. By using the SLIDE algorithm video frames are projected to the coordinate axes. In other words, analysis of a three dimensional signal is reduced to one dimensional cases, along two axes. Beside the basic form of the SLIDE algorithm, based on the constant μ propagation, its modifications are considered. These modifications are obtained by using variable μ propagations. This approach reduces the problem of motion parameter estimation to the instantaneous frequency estimation or instantaneous frequency rate estimation. Namely, instantaneous frequency is related to the velocity, while the instantaneous frequency rate is related to the acceleration. The spectrogram, the S-method and the local polynomial distribution are used for these purposes. Also, the case of active camera system is analysed. A non static background introduces additional difficulties in motion parameter estimation. The theoretical considerations are illustrated by numerical and real examples.