Time-Variant Moving Objects Extraction using Velocity-Tuned Filter Banks and Time-Recursion Method in Infrared Image Sequences

Abstract

A motion vector selective moving object estimation algorithm that preserves the exact shapes and textures of moving objects is presented. In order to extract multiple moving objects with arbitrary motion vectors embedded in the sequence of image frames of cluttered stationary background as alleviating the aliasing effects, both 3D spectral filter banks, called velocity-tuned filter banks, and time-recursive Kalman filter are incorporated to work in parallel. Furthermore, using the fact that the motion energy for each one of the moving objects takes a unique part of the spectrum in the 3D spatio-temporal frequency space, the rotation invariant multiple moving objects detection is also possible when using the proposed filter banks. Simulations have been run to analyze the performance of our filtering algorithm utilizing image sequences of natural scenes. The accurate and robust sets of estimation results are observed down to signal-to-noise ratios of 12 dB.