What image information is important in silhouette-based gait recognition?

Gait recognition without subject cooperation

“…May 24, 2010 DRAFT a silhouette still contains information about the shape of human body that is vulnerable to changes caused by conditions such as clothing and carrying. Although recent studies suggest that static shape information is more important than kinematics for most of the silhouette-based gait recognition approaches [18], [19], including static appearance features in gait representation also makes the existing approaches vulnerable to the changes of covariate conditions. To overcome the problem, it is crucial to select the most relevant gait features that reflect the unique characteristics of gait as a behavioral biometric, and importantly are invariant to appearance variations caused by changes of covariate conditions.…”

“…We argue that the main reason for the poor performance is that the existing approaches rely on both static appearance features and dynamic gait features for person identification, i.e. the identification is not achieved using gait alone [18], [19]. More specifically, most existing approaches represent gait using features extracted from silhouettes.…”

Is Enough Enough? What Is Sufficiency in Biometric Data?

“…it represents subject-dependent covariates. We used our own database as benchmark for convience, since we know its structure and good recognition results were reported using this database [11]. It is assumed that a new database will be recorded in similar or near conditions to the recording of the benchmark and the quality of images in the new database will be not much worse than in the benchmark.…”

Human Perambulation as a Self Calibrating Biometric

“…The widely used and compared databases on gait recognition include: the University of Maryland's surveillance data [10]; the University of South Florida's outdoor data [11]; Carnegie Mellon University's multiview indoor data [12]; and the University of Southampton's data [13]. The majority of methods and databases found in the literature thus concern a person walking in frontoparallel [6], [7], [9] or the use of several digital cameras acquiring the movement [7], [8] and thus the knowledge of the calibration parameters.…”

“…On the other hand, model-free approaches are mainly silhouette-based. The silhouette approaches [8], [9] characterize body movement by the statistics of the patterns produced by walking. These patterns capture both the static and dynamic properties of body shape.…”