Person re-identification using deep siamese network with multi-layer similarity constraints

Person Re-identification is aimed to identify a person through multiple camera views. The task has attained a huge research interest due to its apparent importance in surveillance systems from security aspects. This paper introduces a novel methodology based on Siamese architecture with multi-layer similarity constraints. The baseline model embraces two dense blocks to preserve feature maps at each convolutional layer. Besides, the model...