NTT has demonstrated a monitoring system that selects details from high-resolution camera footage before sending them to a model that interprets images and language. Its preliminary processor tracks objects and passes close-up images with movement information for closer assessment. A company demonstration used a road-accident recording; road and airport monitoring are intended applications, with further development planned.
NTT, the telecommunications group, has demonstrated a monitoring arrangement that combines high-resolution camera footage with a model able to interpret images and language together. Intended uses include roads and airports. NTT’s own AI processor first examines the footage, detecting and tracking objects before deciding which scenes require more detailed interpretation.[1], [2]
In the company’s description, the first stage avoids shrinking the entire 4K image before assessment. It estimates the objects’ location, speed and direction of travel. When a situation calls for closer examination, a high-resolution crop around an object goes to the model with its movement information. The purpose is to retain details of small or distant objects that can disappear when a whole image is reduced, while providing the motion data needed to interpret the surrounding scene.[1]
The company tested the arrangement on a one-minute road-accident recording. A yellow vehicle hits a structure between carriageways and stops across the road. NTT says its initial processing detected the stopped vehicle and the model interpreted the accident from the cropped image and surrounding traffic movements. The footage came from a mobility-research dataset and was upscaled to 4K for the exercise. The result belongs to the company’s demonstration. NTT plans further development for different monitoring settings, including the preliminary layer that selects information for generative AI. NTT Innovative Devices plans engineering samples containing the processor within the fiscal year, and the work is scheduled for display at NTT R&D FORUM from November 18 to 25.[1]
