Construction of metadata of video for effective video search

Video searching in the existing video sharing platform such as YouTube depends on hashtags or basic metadata generated manually when the video is produced.In such systems, the search accuracy is low because there is no detailed information about the video content.In this paper, we implemented an AI-based metadata construction system, the VMeta, which analyses audio and video frames in a video to extract feature data and stores them as 13 useful metadata to increase the accuracy of video search.In VMeta, when a video is uploaded using a web service, it analyses and creates metadata using TensorFlow.The VMeta system provides users with various meta-information about the searched video, such as presenter, keywords, categories, and ten-second text scripts, making it easy for users to make choices and provide users with a variety of information about the video without having to play it.