Surveillance camera image dataset for visual classification in criminal investigation research (CCTV cameras)

Abstract

This data article describes a curated collection of 1421 images of civilian surveillance and security cameras, assembled to support computer vision research on the visual recognition of camera systems in real-world environments. The images were gathered from freely reusable web sources, open image repositories, and manufacturer product pages, and every image was manually reviewed so that each retained file clearly depicts at least one surveillance camera. The collection covers the most common camera form factors, including dome, pan-tilt-zoom (PTZ), bullet, fixed box, multi-sensor, and consumer IP or doorbell cameras, photographed in outdoor street installations, indoor security setups, pole-mounted and wall-mounted configurations, as well as manufacturer product imagery against plain backgrounds. All files follow a sequential naming convention (camera_<ID>) and are provided in a single directory in JPEG format, in RGB colour space, standardised to a uniform resolution of 640 × 640 pixels. A lightweight annotation file (labels.csv) accompanies the images, recording for each file whether it depicts a camera installed in a real environment (718 in-situ photographs) or a manufacturer-style product image (703 images), together with a per-image camera_type label giving the form factor of the device shown. A documented Python loading script and SHA-256 checksums are distributed with the data. The dataset is publicly available on Zenodo under a CC BY 4.0 licence. It can be reused to train and benchmark image classification and object detection models for automated camera canvassing at crime scenes, and to support research in digital forensics, privacy auditing, urban surveillance mapping, and smart-city applications.