This article is cited in 4 papers

NUMERICAL METHODS AND THE BASIS FOR THEIR APPLICATION

On quality of object tracking algorithms

D. A. Vrazhnov^a, A. V. Shapovalov^abc, V. V. Nikolaev<sup>b

a</sup> Tomsklabs LLC, Akademicheskiy av., 8/8, Tomsk, Russia, 634055
^b Tomsk State University, Lenin av., 36, Tomsk, Russia, 634050
^c Tomsk Polytechnic University, Lenin av., 30, Tomsk, Russia, 634050

Abstract: Object movement on a video is classified on the regular (object movement on continuous trajectory) and non-regular (trajectory breaks due to object occlusions by other objects, object jumps and others). In the case of regular object movement a tracker is considered as a dynamical system that enables to use conditions of existence, uniqueness, and stability of the dynamical system solution. This condition is used as the correctness criterion of the tracking process. Also, quantitative criterion for correct mean-shift tracking assessment based on the Lipchitz condition is suggested. Results are generalized for arbitrary tracker.

Keywords: computer vision, tracking, mean-shift, dynamical systems.

UDC: 004.931;004.932

Received: 14.02.2012
Revised: 04.05.2012

DOI: 10.20537/2076-7633-2012-4-2-303-313