Using the spatio-temporal characteristics and spectral properties of motion in the sequence of images a motion estimation algorithm for detecting and estimating the exact shapes and textures and instantaneous positions of moving objects is presented. The described motion analysis filter bank incorporating the motion-vector-selective fan filters with the time-recursive Kalman filter working in parallel has an ability of extracting the multiple moving objects embedded in the sequence of image frames of the cluttered stationary background. Simulations were performed to verify the claimed properties of our filter bank using real, natural terrain image sequences. The results have shown us a good performance of detecting and estimating each one of multiple moving objects that undergoes constant rate translation and rotation in both of noiseless and noisy image sequences.
展开▼
