Here we gather information about people’s perception of safety by asking the users to choose what image is safer from a random pair.
The votes are inserted in a database, alongside the location and voting time, and used to train a neural network to automatically predict a safety score from new images in milliseconds.
Available options are the first, the second, or both are equally safer/dangerous. This crowdsourcing is on the main Website, cycleai.net. There are 5000 points available, half from London and half from Lisbon. Each point has 4 images, corresponding to 0-, 90-, 180- and 270-degrees orientation. The images to be displayed are selected from a random uniform distribution.