Motion estimation from motion smear - a system identification approach

Abstract

Motion smear, which arises because of fast motion relative to the shutter time of a camera, is generally considered as an artifact. Little work has been done to use motion smear as a visual cue for motion estimation or image restoration. Here, we present a new approach to estimate motion from two successive frames of smeared images. The blurring system is modeled as temporal integration of instantaneous images and has been estimated using system identification theory. Motion parameters have been extracted from the estimated system. As compared to earlier approaches having a similar objective, no edge detection or optical flow analysis is required. Our approach establishes a trade off between signal to noise ratio (SNR) and computational complexity. Highly accurate results have been observed with SNR as low as 12 dB. Experimental results with both simulated and real images are shown.