Towards view-invariant gait modeling: Computing view-normalized body part trajectories

Abstract

This paper proposes an approach to compute view-normalized body part trajectories of pedestrians walking on potentially non-linear paths. The proposed approach finds applications in gait modeling, gait biometrics, and in medical gait analysis. Our approach uses the 2D trajectories of both feet and of the head extracted from the tracked silhouettes. On that basis, it computes the apparent walking (sagittal) planes for each detected gait half-cycle. A homography transformation is then computed for each walking plane to make it appear as if walking was observed from a fronto-parallel view. Finally, each homography is applied to head and feet trajectories over each corresponding gait half-cycle. View normalization makes head and feet trajectories appear as if seen from a fronto-parallel viewpoint, which is assumed to be optimal for gait modeling purposes. The proposed approach is fully automatic as it requires neither manual initialization nor camera calibration. An extensive experimental evaluation of the proposed approach confirms the validity of the normalization process.

Estimation of 3D head region using gait motion for surveillance video

“…Then, Explorative Particle Filtering was applied to detect people and for 3D tracking. In another approach, Jean et al [5] used the 2D trajectories of both feet and the head extracted by using the silhouettes. After that, the frontoparallel normalized view trajectory was generated from a homography transformation based on the 3D walking plane.…”

Gait recognition using normalized shadows

“…LDA identifies a projection matrix ∅ onto a subspace that maximizes the ratio of intra-to inter-class scatter, using Fisher's criterion.  Given k classes with centroids ̅ and a test GEI , the system computes the Euclidean distance (, ) in a transformed space, and selects the class with the lowest distance, according to equation (6).…”

“…Examples include the construction of 3D models using information from multiple 2D video cameras [2] or from sensors capturing 2D video and depth information [3]. Other methods model the user's motion using key points such as the hip, knee and ankle positions [4], [5] or the head and feet positions [6]. These recognition methods rely on the use of additional information such as depth, external and internal camera parameters, floor position, etc.…”

Improving Human Gait Recognition Using Feature Selection

“…Model-based methods attempt to explicitly model the human body or motion by employing static and dynamic body parameters, which are typically view and scale invariant. Different approaches have used features like distances or angles between different human body parts [1,8,9], trajectories of joint angles, head or feet [10,11,12] using 2D stick models, motion parameters using 3D temporal models [13], or a combination of kinematics and appearances of a gait [14]. Model-free approaches, on the other hand, usually employ either shape of binary silhouettes or the whole motion of walking person's body, rather than modeling the whole human body or any part of it.…”