Towards pose invariant gait reconstruction

Abstract

Recently, a lot of research has been conducted into the usefulness of gait for identification at a distance. Since the gait of a person is readily identified when extracted from a canonical side view, most algorithms work with the premise that the motion is frontoparallel in nature, or require some knowledge of the camera calibration. Realistically people will always walk along different trajectories to the camera. In this paper we show that gait has sufficient properties that allows us to exploit the structure of articulated motion within single view sequences, in order to remove the unknown subject pose and reconstruct the underlying gait signature, with no prior knowledge of the camera calibration.

Recognizing people in non-intersecting camera views

“…Results on synthesized data showed that simple pose correction for geometric targets generalizes well for objects on the optical axis. More recently, these techniques have been refined and applied on subjects wearing reflective markers with success observed from 6 different point views [15].…”

“…The method of Carter and Nixon corrects the variations in gait data by knowing the walking trajectory and modelling the thigh as a simple pendulum. The approach was then reformulated by Spencer and Carter [15] to provide a pose invariant biometric signature which did not require knowledge of the subject's trajectory. Results on synthesized data showed that simple pose correction for geometric targets generalizes well for objects on the optical axis.…”

“…The method proposed by the authors in [15] is based on four main assumptions: the nature of human gait is cyclic; subjects walk along a straight line; the distances between the bone joints are constant; and the articulated leg motion is approximately planar. Therefore, multiple periods of linear gait motion appear analogous to a single period viewed from many cameras related by linear translation and the positions of the legs joints lie in an auto-epipolar configuration.…”

“…The approach is based on marker-less feature extraction to recover joint positions of walking subject from uncalibrated single cameras. The gait signature is thereafter derived in the sagittal plane by a rectification process [15] that transforms gait angular data from a particular viewpoint to the normal plane. In an unconstrained environment where camera information is not available, people are tracked from different viewpoints by matching their gait signature against a database of existing signatures.…”

Human Perambulation as a Self Calibrating Biometric

“…Recently, in fact, novel approaches on biometrics based on gait are oriented towards synthesizing fronto-parallel views by use of structure from motion, but require some information about camera calibration [14], [15]. Moreover, considering the recent applications of gait recognition in criminal investigation, like the case of the murderer of Swedish Foreign Minister Anna Lindh [16], usually there is no access to the camera and generally only the recorded video sequences are available [17], [18].…”