Character interaction detection method based on bidirectional attention mechanism under knowledge guidance

Abstract

The invention discloses a character interaction detection method based on a bidirectional attention mechanism under knowledge guidance. The character interaction detection method comprises the following five steps of target detection, grouping, bottom-up encoder, top-down encoder and binary classification. The key of character interaction detection on multiple person and object instances existing in a complex scene is that relation modeling analysis is effectively carried out on objects in the scene, and a knowledge-guided top-down attention mechanism is introduced into a character interaction detection task. The model can adaptively search related information in a scene for a given human-object pair so as to carry out interaction behavior discrimination, so that human interaction detection is realized; the method provided by the invention has better detection performance (mAP) and better interpretability.

Description

technical field [0001] The invention relates to character interaction recognition detection in pictures, in particular to a character interaction detection method based on a knowledge-guided bidirectional attention mechanism, which is a character interaction recognition detection method under a two-stage setting. [0002] technical background [0003] In recent years, with the continuous development of computer vision, the task of human-based interaction detection has received extensive attention. The core problem of the human interaction detection task is to understand and detect the relationship between people and their surrounding objects in the scene. This task has many broad application scenarios: it can be directly applied to monitoring systems, robot vision systems, and human-computer interaction. At the same time, it can be used as a basic interface to provide support for a series of downstream complex visual tasks, such as image and video recommendation and retrieval...

AI Technical Summary

Problems solved by technology

In this case, how to conduct customized relationship analysis for different person-person pairs is a difficult problem, and accurately performing character relationship analysis in this scenario is a challenging problem

Images