Pyramid-based estimation of 2-D motion for object tracking

Abstract

An algorithm capable of estimating the image motion of a moving object in real-time is presented. The method is based on the image correlation technique applied to the multi-resolution image of two successive frames. Unlike previous approaches on the intensity correlation, the method estimates the normal motion vector by establishing the correlation along the direction of intensity gradient only. This approach allows one to avoid difficulties often arise due to the multiple matching. Normal flow vectors are used to estimate the actual flow vector (called the full flow vector) by assuming that 2-D motions of the object and the background are constant. This entire process operates uniformly on several spatial resolutions of image sequence. Each resolution is tuned to a specific range of image motion, and the correct resolution is determined by comparing the estimation error which has been accumulated in the course of full flow estimation. The algorithm has been implemented on off-the-shelf vision hardware, as a subsystem for real-time visual tracking.<>