Here you can download the dataset prepared for machine vision traning. The dataset currently contains nearly 2,000 images illustrating ‘signs of surveillance’ and is explicitly created for the purpose of training computer vision systems to recognise places where computer vision systems or other kinds of video surveillance may be occurring. The contents of the dataset are based upon a six year project photographing, collecting and collating these signs in public spaces across the world indicating that CCTV surveillance is operating. These systems, and their associated warning signs increasingly indicated that automated trained machine vision systems such as ANPR (automated number plate recognition) and face recognition, amongst other systems, are being used. The original photographic collection runs to over 2000 images recorded in 15 different countries and includes images from Belgium, Canada, China, Denmark, France, Italy, Germany, Japan, Luxemburg, Netherlands, Portugal, Sweden, UAE, United Kingdom and United States.
The image set is principally designed for studying the problem of unconstrained CCTV sign recognition in the wild. Each image was prepared by hand to create a uniform set aspect ratio and pixel data per image ready for passing to the selected neural network model for training.
The images within the set are divided by originating country and are full colour RGB images in JPEG format at 250×250 pixels size
The images and dataset are ©2019 Daniel Buzzo and are free for non-commercial use with attribution. For commercial or other uses please contact the author at firstname.lastname@example.org.
You can download the first version of the data set as a .zip compresses archive (circa 55Mb) here.
See the full version history and more on this project at https://signsofsurveillance.com/dataset/