Multi-Year Urban Streetlight Image Collection for Visual Monitoring and Drift Analysis (Part 2/2)

This record provides the remaining 4 streetlight nodes from the multi-year Bristol streetlight imagery dataset collected between 2021 and 2025. These nodes complement the 18-node collection published in the companion record (https://doi.org/10.5281/zenodo.17781192), together covering all 22 fixed-angle Raspberry Pi Camera Module V1 nodes deployed along a ~7.2 km urban corridor across Bristol, UK. Each upward-facing camera captured one JPEG image approximately every hour (with a randomized delay up to 600 s), producing a diverse set of day/night imagery under varying weather and seasonal conditions. The 4 nodes included in this record are: RSS-59-C, RSS-8-C, RSS-A-13-C, RSS-A-33-C. Each node is provided as an independent ZIP archive that can be downloaded and extracted separately. These archives contain JPEG images named using unique 32-character alphanumeric identifiers generated at capture time to act as persistent file references. A central metadata file (streetcare-drift-dataset-2021-2025.csv) is included in the companion record and contains metadata for all 22 nodes, including timestamps, node identifiers, GPS coordinates, day/night labels, mean RGB statistics, automated fault detection results, and two visual drift metrics derived from a CNN-VAE latent-space model. This 4-node subset completes the full 22-node dataset and is intended to support research in visual drift detection, anomaly monitoring, environmental inference, unsupervised representation learning, and MLOps retraining policy evaluation. All data are provided in standard JPEG/ZIP formats for ease of inspection and scripting. Note: The central metadata file (streetcare-drift-dataset-2021-2025.csv) is not duplicated here. It is hosted in the main record (Part 1: https://doi.org/10.5281/zenodo.17781192), which compiles metadata for all 22 nodes.