Analysis of Skeletal Shape Trajectories for Person Re-Identification

Abstract

In this paper, we are interested in people re-identification using skeleton information provided by a consumer RGB-D sensor. We perform the modelling and the analysis of human motion by focusing on 3D human joints given by skeletons. In fact, the motion dynamic is modeled by projecting skeleton information on Grassmann manifold. Moreover, in order to define the identity of a test trajectory, we compare it against a labeled trajectory database while using an unsupervised similarity assessment procedure. Indeed, the main contribution of this work resides in the introduced distance that combines temporal information as well as global and local geometrical ones. Realized experiments on standard datasets prove that the proposed method performs accurately even though it does not assume any prior knowledge.

Person Re-Identification from different views based on dynamic linear combination of distances

“…6 shows that the proposed method outperforms the work presented in [8] with value equals to 90.90% vs. 82% at rank 4 and it reaches a value of 100% at rank 5, whereas the work of [8] reaches the value of 100% only at rank 7. However, the work of [8] outperforms the proposed method on rank 1 with value equals to 45.45% vs. 36.36%. Thus, the proposed method records much higher matching rates than the work of [8] and for some other cases the work of [8] outperforms the proposed method.…”

“…Fig. 6 shows that the proposed method outperforms the work presented in [8] with value equals to 90.90% vs. 82% at rank 4 and it reaches a value of 100% at rank 5, whereas the work of [8] reaches the value of 100% only at rank 7. However, the work of [8] outperforms the proposed method on rank 1 with value equals to 45.45% vs. 36.36%.…”

“…Imani et al [15] have exploited RGB, depth and skeleton information from RGB-D sensors in order to exploit complementarity of features extracted by different modalities [39]. Similarly, in [8], re-id has been based on the analysis of skeleton shape trajectories using different distances. However, given the lack of availability of 3D datasets, many studies have tried to extract skeletons from RGB videos.…”

“…On the one hand, the shape of the skeleton at each frame of the se- quence is projected on the Grassmann manifold resulting to a trajectory on the non linear space. In fact, the projection is mainly performed using Singular Value Decomposition (SVD)-based orthogonalization [8]. Given a skeleton sequence S, we propose to model each frame as a point on the Grassmann manifold (a matrix) in order to transform the video on a trajectory T that links different points on the Grassmann manifold.…”

“…More precisely, we propose an accurate method based on measuring the similarity between trajectories for two persons in order to identify a person of interest across different camera views on different locations and at different times. It is worth noting that this paper is an extension of our previous conference paper [8] towards the general framework of re-identifying a given person among a large set of persons over various camera views. The main contribution of the proposed method resides in adopting machine learning, and more precisely random forest, for the dynamic linear combination of distances (Grassmann distance and local distance) according to each person.…”