SARC3D: A New 3D Body Model for People Tracking and Re-identification

Abstract

We propose a new simplified 3D body model (called SARC3D) for surveillance application, which can be created, updated and compared in real-time. People are detected and tracked in each calibrated camera, with their silhouette, appearance, position and orientation extracted and used to place, scale and orientate a 3D body model. For each vertex of the model a signature (color features, reliability and saliency) is computed from 2D appearance images and exploited for matching. This approach achieves robustness against partial occlusions, pose and viewpoint changes. The complete proposal and a full experimental evaluation are presented, using a new benchmark suite and the PETS2009 dataset.

Color Distribution Pattern Metric for Person Reidentification

“…The experimental results of our methods on 3DPeS dataset show the effectiveness of our similarity classification metric learning. a The accuracy of rank 10 in SARC3D is read from Figure 6 in [18]. b The PUR score is calculated using figures read from Figure 6 in [18].…”

“…a The accuracy of rank 10 in SARC3D is read from Figure 6 in [18]. b The PUR score is calculated using figures read from Figure 6 in [18].…”

“…In this section, we evaluate our method on 2 public datasets: 3DPeS dataset [21] and IPR dataset [18]. These 2 datasets are based on real surveillance setup of University of Modena and Reggio Emilia campus.…”

“…According to 3DPeS dataset [21], background pixels ratios of pedestrian images vary from 34% to 89%, with a mean value of 59.4%. In IPR dataset [18], background pixels ratios vary from 36% to 80%, with a mean value of 59%. The average background pixels ratio is about 59.2% in these two datasets, which could bring great interference for feature robustness and classification if they are not filtered out before final classification.…”

“…This shows great and balanced classification performance of methods on both positives and negatives. However, it is hard to determine which kernel is better in the Re-ID ranking problem from Table 1 Thus, in further experiments, we use these methods with 4 kernel types (where the kernel degree for polynomial kernel is 1) to evaluate our feature models capabilities and compare them with other state-of-the-art methods on 2 public datasets: 3DPeS dataset [21] and IPR dataset [18].…”