Thumbnail
Access Restriction
Open

Author Stauffer, Chris ♦ Lee, Lily ♦ Grimson, Eric
Source CiteSeerX
Content type Text
File Format PDF
Subject Domain (in DDC) Computer science, information & general works ♦ Data processing & computer science
Subject Keyword Non-parametric Articulated Pedestrian Representation ♦ Articulated Model ♦ Valid Position ♦ Sparse Representationholdspromise ♦ Gait Recognition Work ♦ Particular Limb ♦ Different Location ♦ Generative Modeling ♦ Image-based Approach ♦ Boltzmann Machine ♦ Computer Vision ♦ Previous Work ♦ Model Initialization ♦ Difficult Problem ♦ General Application ♦ Explicit Model ♦ Valid Silhouette ♦ Functional Group ♦ Basis Vector ♦ Recent Past ♦ Coefficient Value ♦ Particular Location ♦ Coefficient Vector ♦ Gait Work ♦ Observed Silhouette ♦ Activity Recognition ♦ Body Part ♦ Model Specification ♦ Specific Type ♦ Sparse Basis
Abstract The Problem: Modeling highly articulated objects such as people is an extremely difficult problem in computer vision. This work investigates an alternative to the two extremes of using explicit models and using “exemplars.” Rather than fitting an articulated model to a silhouette or treating the silhouettes holistically, this work explains how to derive a sparse basis to represent the observed silhouettes. The coefficients of the basis vectors indicate the presence of a body part at a particular location. Motivation: The advantage of this representation is the ability to exploit structure in the coefficient values. We can factor the basis into “functional groups ” representing particular limbs in different locations. We are investigating how a Boltzmann Machine can be used to constrain coefficient vectors to the manifold of “valid ” silhouettes. This factorable, sparse representationholdspromise for generative modeling and activity recognition. Previous Work: Articulated models have become vastly more useful in the recent past. Unfortunately, model specification, model initialization, and tracking stability have limited their use for general applications. Constraining the models to the manifold of “valid ” positions is difficult, although it has been done for specific types of actions [2]. Image-based approaches to articulated tracking have been severely limited. Using correlation between silhouettes, some gait work has been accomplished[3]. Other gait recognition work have used descriptions of motion in
Educational Role Student ♦ Teacher
Age Range above 22 year
Educational Use Research
Education Level UG and PG ♦ Career/Technical Study