Efficient 3-D filter banks for motion estimation (Q2784724)

The aim of the paper is to propose an improved approach to the problem of object tracking. The authors consider a velocity selective filtering approach based on spatio-temporal filter banks. These filter banks represent an extension of the 2-D directional filter bank technique for emulating the behaviour of the spatio-temporal continuous wavelet transform (CWT). The CWT algorithm proved to be useful to estimate motion parameters corresponding to moving objects in a 3-D sequence, and it was shown to perform well in noisy environments and with multiple objects present. But the CWT technique relies on an over-sampled representation of the wavelets, which is computationally inefficient. The proposed directional filter bank approach to the object tracking algorithm is based on 3-D matched filtering and offers a computational advantage over direct implementation of the CWT-based filters. The spatio-temporal directional filter banks exploit the temporal information, reduce the computational complexity by a factor of two over the conventional approach, and provide additional flexibility and adaptivity.NEWLINENEWLINEFor the entire collection see [Zbl 0971.00016].