Object Tracking and Image Restoration in the Dark

Abstract—We show a method of estimating object motion and original image simultaneously using a Bayesian framework to realize object tracking and image restoration in the dark. The motion parameters which maximize marginal likelihood when input images are observed are estimated. In finding a solution using an iterative method, a broader search is performed by calculating differences after applying a strong low-pass filter to the input images. As a result, we realized object tracking and image restoration from artificially generated video images with an SNR of up to −6 dB. We also showed effectiveness of the proposed method by comparing it to a conventional method using the same low-pass filter. Moreover, we conducted an experiment using real video images captured in the office at night with all the lights off, and object tracking with rotary motion and image restoration with color information were successfully realized. Keywords—Image restoration, denoising, Bayesian estimation, merginalization, multi-frame image processin