Sitewatcher

Over the last 10-15 years, many security platforms have been put in place to capture security images for monitored sites, but almost all of those lack any smart capability to recognise the subjects in those images. This means rudimentary methods, such as motion sensing, continues to be the primary method for triggering security alerts and therein a recipe for more false alarms than useful and accurate ones.

At the same time, these connected cameras represent a large sunk cost for most businesses and cannot easily be updated to modern ones that may afford smart recognition capabilties. Ideally, a solution would be found to allow these older (dumb) camera networks to be modernised without needing to replace them.

Using Google's cloud AI and ML micro-services, we were able to build an site monitoring cloud application (Sitewatcher) that examines each triggered site photo. It looks at the content of each photo, specifically looking to classify the content of the image by objects identified (such as human, car, license plate, etc.) and then using that classification and a powerful set of context-based rules, it determines whether this is a valid trigger event worthy of escalation.

Using context, allows our rules engine to reinforce its initial assessment with things like:

🔸 multiple similar events occuring within a feasible amount of time (for example, ignoring multiple unrecognised visual artefacts that are unclassifiable, as likely lighting anomalies), or

🔸 within a defined known set of parameters (for example, a car detected may be ignored if the colour and license plate are within an approved list)

This also allows the recognition system to improve its accuracy over time, for each camera and the camera's image quality, making it more likely that a poor quality image of a human would still be recognised for that poor quality camera.

Sitewatcher allows these older security camera fleets to continue to be used, but with added smart capabilities derived through using easily accessible cloud services.

The application excels at identifying humans, and other objects in sub-standard images, and then adds AI decision-making smarts to understand the context and known site parameters, before activating any alarms or escalation workflows.