Tracking feature extraction based on manifold learning framework

摘要

Manifold learning is a fast growing area of research recently. The main purposenof manifold learning is to search for intrinsic variables underlying high-ndimensional inputs which lie on or are close to a low-dimensional manifold.nDifferent from current theoretical works and applications of manifold learningnapproaches, in our work manifold learning framework is transferred to trackingnfeature extraction for the first time. The contributions of this article include threenaspects. Firstly, in this article, we focus on tracking feature extraction forndynamic visual tracking on dynamic systems. The feature extracted in this articlenis based on manifold learning framework and is particular for dynamic trackingnpurpose. It can be directly applied to system control of dynamic systems. This isndifferent from most traditional tracking features which are used for recognitionnand detection. Secondly, the proposed tracking feature extraction method hasnbeen successfully applied to three different dynamic systems: dynamic robotnsystem, intelligent vehicle system and aircraft visual navigation system. Thirdly,nexperimental results have proven the validity of the tracking method based onnmanifold learning framework. Particularly, in the tracking experiments the visionnsystem is dynamic. The tracking method is also compared with the well-knownnmean-shift tracking method, and tracking results have shown that our methodnoutperforms the latter.