and Dmitry Chetverikov
Image and Pattern Analysis Group
Computer and Automation Research Institute
Budapest, Kende u.13-17, H-1111 HUNGARY
We present the current results of an ongoing systematic performance evaluation study of feature point tracking algorithms [11,7,10,9,4]. After a description of the problem and the alternative approaches, we share the experience we gained while testing the algorithms. Algorithmic features important in vision applications are discussed. Applications and techniques differ in admissible events, character of motion, and merits of tracking performance. We offer guidelines to selection of a tracking technique by considering the algorithms' scope, error rate, processing time, as well as robustness to non-admissible events, point density and speed. The algorithms are available for testing over the Internet.