On the registration of an object translating on a static background

摘要

This work considers the determination of the relative movements between consecutive images, taken from the same scene (registration), a problem arising in many application fields. With the exception of specialized techniques for specific applications, the most reliable registration algorithms are based on the use of 2-D Fourier transforms. They typically require a relevant computational effort. The computational bulkiness typical of such methods can be overcome by 1-D techniques, needless to say, at the price of a degraded performance in presence of disturbances. This paper proposes a new 1-D method, which exploits the availability of several frames, typical of practical contexts, in order to determine the translation of an object moving between two endpoints on a static background. The noise performance of the new technique is shown to be quite superior to that of the reported 1-D methods. For sufficiently long sequences the noise performance favourably compares with that of 2-D techniques, with considerable computational savings. The possibilities of the method are exemplified by a number of practical examples.