Person Re-Identification

The task of Person Re-Identification is the task of matching images of a person across multiple camera views and identifying if the same person is present in all of them or not. However, this throws up three main challenges:

• Viewpoint Change

• Illumination Variation

• Partial Occlusion

In this work, we propose a Siamese Convolutional Neural Network architecture to tackle these challenges, using a soft-matching approach based on Normalized Cross-Correlation. This model however, suffers from occasional false matches. We therefore, propose a second deep neural network architecture (shown in the figure above) to address the false matching aspect as well. This neural network essentially incorporates both an exact and an inexact matching technique to do this.

Code:

Link

Publication:

• A. Subhramaniam, M. Chatterjee, A. Mittal, “Deep Neural Network with Inexact Matching for Person Re-Identification”, Advances in Neural Information Processing Systems 2016 (NIPS 2016).