Velocity estimation and moving object's detection by using wavelet transform and parallel Kalman filter

We propose a novel method for velocity estimation and detection of a moving object in an image sequence. This purpose is achieved by using discrete wavelet transform and parallel bank of extended complex Kalman filter in the transform/spatio-temporal mixed-domain. In the mixed-domain, image sequence processing is replaced by l-dimensional(l-D) complex signal processing. Then, trajectory signal with approximately constant speed and orientation is considered as a bunch of 1-D complex sinusoidal waves. By applying extended complex Kalman filter to each 1-D complex signal, object's velocity can be estimated. In addition, these estimates can be combined with depending on the energy of the subband images. Through some simulation results, it is shown that moving object's velocity is accurately estimated and the object is detected effectively.