This paper provides a comprehensive survey of person re-identification (re-ID), a task that aims to identify a person of interest in different cameras. The authors trace the history of re-ID, from early hand-crafted algorithms to modern deep learning systems, and classify current methods into image-based and video-based categories. They review both hand-crafted and deep learning systems, highlighting recent advancements such as end-to-end re-ID and fast re-ID in very large galleries. The paper also discusses critical future directions and under-developed issues in the field. Key contributions include: 1. **History and Relationship**: Introduces the evolution of re-ID from multi-camera tracking to independent re-ID tasks, and its relationship with image classification and instance retrieval. 2. **Image-Based Re-ID**: Reviews hand-crafted and deep learning systems, focusing on pedestrian descriptions, distance metric learning, and datasets like VIPeR, CUHK01, and Market-1501. 3. **Video-Based Re-ID**: Explores hand-crafted and deep learning methods, emphasizing multi-shot matching, temporal information, and datasets such as ETH, PRID-2011, and iLIDS-VID. 4. **Future Directions**: Discusses emerging trends like end-to-end re-ID and large-scale re-ID, and highlights the need for larger datasets and more efficient retrieval methods. The paper concludes by emphasizing the potential for significant advancements in re-ID, particularly in handling larger datasets and improving recall rates.This paper provides a comprehensive survey of person re-identification (re-ID), a task that aims to identify a person of interest in different cameras. The authors trace the history of re-ID, from early hand-crafted algorithms to modern deep learning systems, and classify current methods into image-based and video-based categories. They review both hand-crafted and deep learning systems, highlighting recent advancements such as end-to-end re-ID and fast re-ID in very large galleries. The paper also discusses critical future directions and under-developed issues in the field. Key contributions include: 1. **History and Relationship**: Introduces the evolution of re-ID from multi-camera tracking to independent re-ID tasks, and its relationship with image classification and instance retrieval. 2. **Image-Based Re-ID**: Reviews hand-crafted and deep learning systems, focusing on pedestrian descriptions, distance metric learning, and datasets like VIPeR, CUHK01, and Market-1501. 3. **Video-Based Re-ID**: Explores hand-crafted and deep learning methods, emphasizing multi-shot matching, temporal information, and datasets such as ETH, PRID-2011, and iLIDS-VID. 4. **Future Directions**: Discusses emerging trends like end-to-end re-ID and large-scale re-ID, and highlights the need for larger datasets and more efficient retrieval methods. The paper concludes by emphasizing the potential for significant advancements in re-ID, particularly in handling larger datasets and improving recall rates.