Scap net: A capsule network based approach for person re-identification

Abstract

Automatic individual re-identification proof in a multi-camera observation arrangement is significant for powerful following and checking swarm development. As of late, a couple of deep learning-based re-ID methodologies have been created which are precise be that as it may, time-escalated, and in this way unacceptable for reasonable applications. Capsule networks have demonstrated empowering results on de facto benchmark computer vision datasets, for example, MNIST, CIFAR, and smallNORB. This paper proposes a SCap Net, a deep network architecture for person re-identification. With this SCap Net, we present Siamese Capsule networks, another variation that can be utilized for pairwise learning errands. The model is prepared to utilize contrastive loss with l2-normalized capsules encoded pose features. The proposed technique has been assessed on three publicly available datasets: VIPeR, CUHK01, and CUHK03. Results obtained are presented in the CMC curves show that our methodology resolves the orientation and pose based dissimilar image problem and also improves the Rank 1 accuracy from various existing approaches. © Springer Nature Singapore Pte Ltd. 2020.