NEC : technology instantly detects target actions from live video footage by analyzing a few seconds of sample video, eliminating the need for time-consuming AI training

This technology makes it possible to detect actions and movements of individuals who may require assistance and help, such as crouching or falling, in real time by analyzing live video streams, which can enable facility managers to promptly respond to an incident and take effective emergency measures.

With traditional technologies, in order to automatically detect a specific action from video footage, it is necessary to prepare a large amount of video that displays a target action and to have artificial intelligence (AI) learn from the footage. Since it takes time to train AI in this way, it has been difficult to immediately register and detect target actions. In addition, since there are a wide range of variations for each action, it has been difficult to define actions and to train AI to detect them.

Unlike existing video retrieval technologies, NEC's On-Demand Action Detection eliminates the need for preliminarily defining target actions and collecting large amounts of training data for AI. Also, since the new technology can detect similar actions regardless of a subject's orientation, there is no need to analyze multiple sample videos to detect a single action. This significantly reduces the burden on system operators and paves the way for more practical action detection services.

Going forward, NEC will proceed with trails of this technology, aiming to commercially launch applications that include video monitoring services and video content editing services.