Attention model method of video surveillance pedestrian search based on natural language description

Abstract

An attention model method for video surveillance pedestrian search based on natural language description is provided. The attention weights formed by fused feature vectors are used to weight the imagefeature vectors to obtain the degree of association between text and image. At the same time, the text is weighted by the attention weights formed by the text feature vectors, and the importance of the text is obtained. Attention evaluation score is obtained by multiplying the degree of connection between text and image by the importance of the text. The invention uses three different attention weighting modes, more effectively highlights the importance of each word text vector and improves the representativeness and effectiveness of the feature vector, so that the feature vector can be usedmore fully and effectively. The invention can make full use of the effective features in eigenvectors to improve the training and testing effect. The invention can solve some related problems such asartificial intelligence, machine learning and video surveillance pedestrian search based on natural language description.

Description

technical field [0001] The invention relates to an attention model method for pedestrian search in video surveillance. In particular, it concerns an attention model approach for pedestrian search in video surveillance based on natural language description. Background technique [0002] With the continuous development of artificial intelligence and deep learning, people's requirements for artificial intelligence and deep learning are also getting higher and higher. The feature vectors extracted from images and texts and the effective use of these feature vectors are very important for the training and testing process of deep learning, and can even play a decisive role. In order to enable deep learning to use feature vectors more effectively, many attention models have been proposed and widely used in various types of deep learning tasks, such as image classification, image recognition and subtitle generation. Meanwhile, different kinds of attention models are also frequentl...

AI Technical Summary

Problems solved by technology

[0003] Compared with the widely used attribute-based pedestrian search, video surveillance pedestrian search using human natural language description is more conducive to the use of actual scenes, but the difficulty of implementation is also higher than attribute-based pedestrian search

In the video surveillance pedestrian search task based on human natural language description, the effect of the existing attention model algorithm is not ideal

Images