AI is being used to help identify and rescue victims of human trafficking.
The technology, developed by researchers from George Washington University, Temple University and Adobe, initially launched an app called TraffickCam in 2016. The app, which allows people to take photos of their hotel rooms and submit them via the app, has amassed one million images from over 50,000 hotels worldwide. The photographs are then put through an AI engine, a deep convolutional neural network, which learns a set of filters enabling it to decipher which hotel the photo could have been taken at. Décor such as curtains, wallpaper, and bed linen can be analysed to narrow down the hotel to particular chains or locations.
The Hotels-50K dataset is intended to be used to match up with online adverts placed by traffickers, who use selfies taken by their victims in hotel rooms. The amateur photographs uploaded to the TraffickCam app are thought to be particularly valuable since the quality will likely emulate that of the selfies taken by the sex trafficking victim. However, professional photoshopped images of the hotels, such as you would find on travel websites, also make up the dataset.
Two pre-trained neural networks (ResNet-50 and VCG) have been used to test the dataset. Both could correctly identify common hotel chains from images with almost 80% accuracy. Identifying the individual hotel has been more difficult, though, with the system finding the correct hotel within the first 100 images only 24% of the time.
Results have not yet been released on how effective the system is with real human trafficking photos or if the system has lead to the recovery of any trafficking victims. However, the search system is being used by the National Center for Missing and Exploited Children (NCMEC), a US non-profit organization.
The International Labor Organization estimates that there are 4.5 million people in sexual slavery around the world.