Motion Blur-Based State Estimation

Abstract

Motion measurement increasingly deploys image sensors such as charge-coupled device and CMOS arrays, driven by their ever-improving resolution, response time, noise level, and cost. The typical usage is to operate an image sensor and the associated optics as a sampler, by taking a series of high-speed sharp pictures to infer motion. Image blur is treated as an undesirable artifact, to be removed using shorter exposure times or image processing techniques such as deblurring. We have previously shown that dynamic information embedded in image blur may be exploited for model identification in frequency ranges well beyond the Nyquist frequency. In this brief, we investigate the state estimation problem using motion blur. We pose the problem as a minimization, estimating the state at the start of each (slow) sampling period based on the observed motion blur. We show that the local convexity of the minimization corresponds to a generalized observability criterion. This method is compared with other techniques, including the conventional centroid-based method, and that based on the use of multiple image moments. The simulation and experimental results demonstrate the fast response and robustness of the proposed scheme in the presence of synthetic stray light.