One challenge in image or video analysis for law enforcement is understanding their spatio-temporal context. This means not just figuring out what the image is about, but also answering questions like:

• When and where was this material created?

• Is this a part of a series?

• Have we seen similar material from this place or time before?

To do this, we need to analyse the content of the image or video itself, rather than relying on metadata which can often be missing, obscured or incorrect. This project is focused on developing new methods for identifying when and where photographs and videos were taken, based purely on the content and objects in the images themselves.

It leverages the StreetCLIP model, which builds upon the open-sourced OpenAI CLIP (Contrastive Language-Image Pretraining) model by using open-sourced Google Street View imagery datasets. The project is re-training this by testing the use of scraped real estate data for further detection.

This will enable us to go beyond just detecting features of an image or video to extracting more details, which can give us the likely suburb, city and country an image or video was taken from – supporting search-and-rescue efforts and crime scene analysis.