Background suppression for building accurate appearance models in human motion tracking
MetadataShow full item record
This paper presents a robust and fully-automatic human motion tracking system without motion priors information using a camera in a fixed location. Bottom-up estimation approaches have recently been applied to such tasks with some success. However, the performance of these approaches is limited by the difficulty of building an effective appearance model. In particular, the appearance model must be derived from initial estimates of the tracked person's limb posture. However, in addition to inaccuracies in this initial estimate, the precise shape, size and boundaries of the tracked person's limbs are not known. Hence it is inevitable that background (non-limb) pixels are included into the appearance model. In the case of smaller limbs such as the arms, this can cause the model to become unrepresentative and sometimes confused with other body parts such as the torso. In this paper, we address the problem of how to automatically extract accurate training samples for building an accurate appearance model, and propose a mechanism for identifying and removing background (negative) pixels via pixel clustering that is robust even with a loose-fitting body shape model. Experiments are conducted to compare the proposed approach against existing appearance-based algorithms without negative pixel removal using several publicly available data sets. Results show that tracking accuracy is consistently improved, and significantly so for small limbs such as the arms.
Showing items related by title, author, creator and subject.
Kramer, Annika (2009)Visual perception is our most important sense which enables us to detect and recognise objects even in low detail video scenes. While humans are able to perform such object detection and recognition tasks reliably, most ...
Leoputra, Wilson Suryajaya (2009)Foreground object detection is a fundamental task in computer vision with many applications in areas such as object tracking, event identification, and behavior analysis. Most conventional foreground object detection ...
Tian, J.; Li, Ling; Liu, Wan-Quan (2014)In this paper we address the problem of tracking human poses in multiple perspective scales in 2D monocular images/videos. In most state-of-the-art 2D tracking approaches, the issue of scale variation is rarely discussed. ...