AI is good at recognizing objects in a photo or video stream. Leading AI engines can now recognize people, cars, dogs, trees, and a whole host of objects. In collaboration with one of our retail clients, we’ve been training an AI engine to recognize weapons: handguns, rifles, and shotguns. We’re also training the engine to spot violent movements and people wearing “ski mask” baklavas, intending to reduce the risk of violence and dangerous experiences.
AI image detection works by submitting a large number of ‘training data’ images to a set of algorithms. By telling the AI engine which images have the target object, in this case, guns, the engine will look for patterns. When successful identifications are scored higher, the engine will tweak the variables, repeat the exercise, and after several “epochs,” the AI engine will have “learned” how to spot a gun in an image.
So far, using a large set of training imagery and some computing power for the AI engine has been reasonably straightforward. The tricky part is taking the image recognition and deploying it to a real-world situation.
The steps for this are as follows:
This real-world deployment's most challenging part is getting a reliable and consistent video feed. Working with our global client, we are standardizing flash camera use and edge server hardware configuration that can be deployed easily or is already onsite for other tracking purposes.
As we advance, we intend to make this application widely available. We are actively seeking large organizations to help bring this application, standardized camera configurations, and alert functionalities to schools, shops, churches, and campuses. Please contact us below to learn how we can support your organization with our technology.