syntax-based method of providing object classification in compressed video by use of neural network which is learned by cooperation with an external commercial classifier

摘要

The present invention relates to a technique for effectively classifying objects from a normal compressed video such as an H.264 AVC or H.265 HEVC video. More specifically, the present invention relates to a technique for quickly acquiring an object classification result by using syntax information (for example, motion vectors and coding types) obtained by parsing compressed image data to extract an area, that is, a moving object area, in which a meaningful motion is present in a video and then, inputting moving attribute information of the moving object area to a neural network, without recognizing and classifying objects from a compressed video generated by, for example, a CCTV camera through complicated image processing as in the conventional art. In particular, the present invention relates to a technique capable of increasing the reliability of the object classification result by enabling the neural network to learn a large amount of data acquired from various photography environments at high speeds through software processing without human intervention by acquiring a combination of the moving attribute information and the object classification information from a plurality of moving object areas in a manner of transmitting an image of the plurality of moving object areas extracted from the compressed video to an external commercial classifier (for example, an Azure cloud service) and receiving the object classification information, and then, by making the neural network learn the same by using the combination.