Person Re-Identification (ReID) is a critical research domain that uses cross-camera surveillance footage to identify pedestrians for social security and education. Initial ReID methods relied heavily on supervised deep network models, but the large data and manual annotation requirements have highlighted the need for more scalable solutions. This paper explores unsupervised domain adaptation and fully unsupervised learning methods for ReID to address these issues. Clustering algorithms, memory banks, and contrastive loss functions make them robust, but pseudo-label noise makes it difficult to learn accurate feature representations due to weak feature image representation. To overcome these limitations, we introduce a novel Multi-View Embedding Model that captures multiple views of an image to improve feature robustness and discrimination. We also propose a diversity loss function to aid multi-view representation learning. Our method improves unsupervised person ReID performance on Market-1501 and MSMT17, according to extensive experiments.