Probabilistic Multi-Person Tracking with Relative Position Measurements
Year: 2009
Type: article
Abstract: This paper presents a particle filtering framework for tracking multiple persons with a monocular camera. So far, most of techniques based on particle filtering rely on an assumption that measurements on pose and speed of moving persons are sufficiently precise. Unfortunately, such an assumption is often violated due to measurement noises as well as irregular movements of persons. To deal with the problem, we have developed a technique for measuring relative position between persons using occlusion reasoning. In particular, we show how the use of relative position measurements can improve the tracking performance, even in difficult situations where two persons frequently overlap in images.
Authors: Kanji Tanaka, Mamoru Minami
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W42428579
