Filtering image sequences from a moving object and the edge detection problem

this paper, we consider the problem of filtering sequences of images taken from moving objects, with the aim of recovering information about the object itself as well as its underlying motion. We first provide a formal description of the admissible classes of images and (possibly nonrigid) motions, and of the functional relationship between the original image and the observed one (blurring and noisy effects). We then focus on the problem of edge detection, assuming full information about the motion. We propose a procedure that includes a preliminary preprocessing of the measured image, aimed to localize the detection problem and to improve the signal-to-noise ratio. Then, the edge identification is accomplished by an algorithm which implements recursive linear quadratic estimation and hypothesis testing. Finally, the procedure is tested against simulated and real data.