Method for tracking multiple image objects based on a penalty graph for avoiding hijacking of a tracker

Abstract

A method for tracking multiple image objects, includes resampling particles from each of the image objects tracked in a previous image, calculating respective weights of the resampled particles, and predicting and tracking locations of the image objects in a current image based on values obtained by multiplying the locations of the particles by corresponding weights, wherein when an image object of the previous image has a neighboring image object located within a threshold distance, weights of the particles sampled from the image object are calculated by multiplying image values of the current image by penalties, which are values of distances between the corresponding particles and the neighboring image object. The method is capable of avoiding a hijacking problem in which identifiers of nearby objects are confused with each other or disappear during the tracking of multiple objects.

Dependencies are inferred from the claim text.