Localization of a Moving Object by a Moving Observer

Peter Nordlund

Tech. Rep. ISRN KTH/NA/P--94/19--SE, June 1994.

Abstract

We present an algorithm which, using a sequence of images in time, indicates the location of a moving object. The algorithm indicates the object localization after only 2 frames, but the result improves over time. Experimental results, from programs implemented in C++ are presented. The method exploits the brightness constancy constraint in conjunction with an affine fit between 2 consecutive images. After the affine fit has been made, a difference image is produced. A centroid of this difference image is computed. Thnadantroid iCVAP/ndicates the location of the object. This method handles the case with both a moving camera and a moving object. The algorithm is noniterative, computationally cheap, and suitable for an implementation on a pipeline processing machine, (e.g. Datacube MV200). The running time complexity is O(n) for an image containing n pixels.

Full paper: (PostScript 1016k)


Peter Nordlund <petern@bion.kth.se>