3-D Object Tracking with the Adaptive Hyperplane Approach Using SIFT Models for Initialization

Object tracking is still a challenging task, especially if it is d o ne in a realistic env iro nm ent. T he o ngo ing increase o f co m pu tatio nal po w er and the efficiency o f the algo rithm s allo w real-tim e estim atio n o f the o bject’s po se in six d egrees o f freed o m . One o f these algo rithm s is the 3 -D hyperplane appro ach, w hich is u sed thro u gho u t this paper, as it has been pro ven to be fast and accu rate. W e sho w ho w to enhance its ro bu stness by u sing a linear illu m inatio n m o d el to gain m o re insensitiv ity to variatio ns o f the illu m inatio n co nd itio ns. W e also present an ad aptio n to co m pensate appearence changes in case o f ex ternal ro tatio ns. A ltho u gh so m e “ six d egrees o f freed o m ” trackers have been established , the necessary initializatio n is o ften igno red o r is o nly so lv ed ru d im entarily. In co ntrast to this, w e sho w ho w to u se a 3 -D S IF T o bject m o d el fo r initializatio n o f the w ho le tracking system and pro ve its efficiency by experim ental resu lts u sing real im age seq u ences.