Multi-Year Urban Streetlight Image Collection for Visual Monitoring and Drift Analysis (Part 1/2)

This record contains a multi-year streetlight imagery dataset collected by 18 of the 22 fixed-angle Raspberry Pi Camera Module V1 nodes deployed on public lampposts across Bristol, UK, covering a ~7.2 km corridor (Coldharbour Lane, Filton Road, the A4174 ring road, and the UWE Frenchay campus). Each upward-facing camera captured approximately one JPEG image per hour (with a randomized delay up to 600 s) from 2021–2025, yielding 526,096 images under diverse day/night, seasonal, and weather conditions. The 18 nodes included in this record are: RSE-6-C, RSE-16-C, RSE-19-C, RSE-21-C, RSE-22-C, RSE-24-C, RSE-25-C,RSE-34-C, RSE-44-C, RSE-A-7-C, RSE-A-10-C, RSE-A-11-C, RSE-A-15-C,RSS-12-C, RSS-15-C, RSS-47-C, RSS-58-C, RSS-A-2-C. Each node’s data is provided as an independent ZIP archive (e.g., RSE-16-C.zip) that can be downloaded and extracted separately. A central metadata file, streetcare-drift-dataset-2021-2025.csv, is included and contains metadata for all 22 nodes, including: unique image identifier (image_name, a stable 32-character hash) serial (node ID), timestamp, hostname, GPS (lat, lon) time_tag (e.g., 2023-Q1), daynight label fault_detected and confidence mean RGB statistics two drift metrics derived from a CNN-VAE model: relative_centroid_drift (normalized latent distance to the Q1 baseline) relative_recon_error (normalized reconstruction error vs. baseline) Dataset totals: 270,267 daytime images and 255,829 nighttime images, comprising 258,527 “Type-0” clear-view images and 267,569 “Type-1” occluded/challenging images across all nodes. This dataset supports research in visual drift detection, anomaly/on–off monitoring, weather inference, unsupervised representation learning, and MLOps retraining policy design. All data are provided in standard JPEG/CSV formats for easy inspection and scripting. The remaining 4 nodes (RSS-59-C, RSS-8-C, RSS-A-13-C, RSS-A-33-C) are hosted in a companion record:https://doi.org/10.5281/zenodo.17859120