Research

1. An Ensemble of Invariant Features for Person Re-Identification

This paper proposes an ensemble of invariant features (EIFs), which can properly handle the variations of color difference and human poses/viewpoints for matching pedestrian images observed in different cameras with non-overlapping field of views. Our proposed method is a direct re-identification method, which requires no prior domain learning based on prelabeled corresponding training data. The novel features consist of the holistic and region-based features. The holistic features are extracted by using a publicly available pre-trained deep convolutional neural network (DCNN) used in generic object classification. In contrast, the region-based features are extracted based on our proposed two-way Gaussian Mixture Model fitting (2WGMMF), which overcomes the self-occlusion and pose variations.

Figure 1-1. Overview of the proposed framework.

Figure 1-2. Image and mask examples from 3DPeS dataset.

(a)

(b)

(c)

Figure 1-3. VIPeR dataset. (a) Comparison of the state-of-the-art single set of features. (b) Comparison of DCNN, 2WGMMF and their combinations. (c) Comparison to Ensemble, SDALF, ColorInv, eBiCov and EIFold.

(a) 1vs1

(b) 3vs1

(c) 5vs1

(d) 3vs3

Figure 1-4. Comparison to the SDALF, Ensemble, SARC3D, ColorInv and EIFold on the 3DPeS dataset (NvsM: N gallery shots vs M probe shots).


2. Combined Estimation of Camera Link Models for Human Tracking across Nonoverlapping Cameras

We employ a novel approach of combining multiple camera links and building bidirectional transition time distribution in the process of estimation. Through the unsupervised scheme, the system builds several camera link models simultaneously for the camera network that has multi-path in presence of the outliers. Our proposed method decreases incorrect correspondences and results in more accurate camera link model for higher tracking accuracy.

Figure 2-1. A multiple camera tracking system.

Figure 2-2. Camera deployment. Red ellipses are exit/entry zones for four links and the number in blue rectangle denotes the camera number.

Figure 2-3. Asymmetric transition time distributions of all 8 links.

Table 2-1. Tracking accuracy

Figure 2-4. Camera link relationships of the above demo video.