Play. Pause. Rewind. “Was that guy here before? What is he holding…I can’t tell!” If you’ve ever said this to yourself while looking through your surveillance camera’s video archive, things will soon get easier for you.
Scientists from UCLA have developed a camera system that analyzes the live video stream of a surveillance camera in order to generate text about what’s going on in the shot. It can spot an object moving from one scene to another and can even detect if a specific object leaves the shot and returns, which is helpful in spotting suspicious activities.
The system is called I2T, short for “Image to Text”, and the way it works is that a group of computer vision algorithms analyze what’s going on in the shot by consulting a database of over two million images grouped in 500 categories of objects. The program grabs a video frame, recognizes and isolates the main objects in the shot while ignoring unneeded background information in the scene.
It then delivers a text description of what’s going on in the shot, which makes searching for relevant moments in the history of a surveillance camera easier by using keywords that would look for a specific moment.
The technology isn’t advanced or intelligent enough yet to be used in a real life situation, but sets the foundations for a future in smart cameras that can analyze what’s going on.
(Via Fast Company)