A Video Description Method Based on Object Attribute Relationship Graph

Abstract

The invention discloses a video description method based on an object attribute relation graph. The method divides a given long video into video segments that are basically consistent in content, and then analyzes the video content based on each video segment to extract The key objects in the video scene, the key objects extracted here are rectangular boxes containing key objects; each key object extracted is used as a node representing the attribute relationship graph of the object corresponding to the video segment. Perform spatial and temporal feature analysis to describe its attributes and use it as a node attribute; take the relative positional relationship between the acquired key objects as the relationship between the two nodes in the object attribute relationship graph corresponding to the video segment. Connect the edges, and use the relative direction, distance, and relative position relationship between two key objects to change in time as edge attributes. The method achieves an effective and refined representation of complex video content within a certain time, and has good scalability.

Description

technical field [0001] The invention relates to the field of image processing, in particular to a video description method based on an object attribute relationship graph. Background technique [0002] At present, video cameras scattered in every corner of the city constitute a huge visual perception network, which provides an important data source for the city security system. In this context, the demand for applications such as content-based video retrieval, object localization and tracking, and object behavior analysis has greatly increased. By constructing a computer program to automatically analyze the video, extract the features and motion trajectories of the "interesting" objects in the scene content, describe the semantic content of the video, and construct the content summary of the visual scene, which is the urgent need for the computer vision perception network to realize perception It will also provide new representation methods and efficient technical means for...

AI Technical Summary

Problems solved by technology

However, the feature description method based on deep learning requires a large amount of labeled data, and the extracted features still stay at the primary visual level and do not have semantic perception information.

Images