Research Detection & Recognition

Person Re-identification using Distance Learning by Convolutional Neural Networks

We propose a distance learning method using Convolutional Neural Networks to re-identify people based on moving image data obtained from multiple cameras that do not share the field of view. We directly learn embedding vectors in the feature space with CNN as image data.The similarity index between people is obtained by the Euclidean distance between embedding vectors after feature extraction by network.The By using Triplet Loss which simultaneously takes into consideration the distance relation of the same person and the embedded vector of different person as the loss function, learning of the embedding vector which can distinguish the person more accurately is done.

Person Re-identification using Distance Learning by Convolutional Neural Networks