Visual Social Signals for Shoplifting Prediction

Retail shoplifting is one of the most prevalent forms of theft, estimated to cost UK retailers over £1 billion in 2018. One security measure used to discourage shoplifting is surveillance cameras. However, evidence shows that unless these cameras are constantly monitored, they are ineffective. Automated approaches for detecting suspicious behaviour have proven effective but lack the transparency to enable them to be used ethically in real life scenarios. One way to overcome these problems is through the use of social signals. These are observable behaviours which can be used to predict an individual’s future behaviour. To this end we have developed a set of 15 visual attributes which can be used for shoplifting prediction. We then demonstrate the effectiveness of these attributes by deriving a new dataset of visual social signals attributes by manually annotating videos from the University of central Florida Crimes dataset.