Viewpoint tracking method based on geometrical reconstruction and semantic integration

Abstract

The invention discloses a viewpoint tracking method based on geometrical reconstruction and semantic integration. The viewpoint tracking method is characterized by comprising the steps of 1, establishing the visual characteristic set of subgraphs, 2, integrating semantic characteristics, 3, active learning algorithm, 4, optimization updating, and 5, ranking and orderly linking the subgraphs to obtain a viewpoint tracking path. The viewpoint tracking method is capable of quickly and accurately detecting various remarkable regions of images, improving the accuracy of viewpoint tracking and then improving the prediction capacity of a viewpoint transfer route.

Description

technical field [0001] The invention belongs to the technical fields of image cognitive reconstruction, image enhancement, and image classification, and mainly relates to a viewpoint tracking method based on geometric reconstruction and semantic fusion. Background technique [0002] Viewpoint tracking is an intelligent image analysis method. Its purpose is to quickly find the visual information that users are most interested in and explain complex scenes by imitating human vision transfer. It is one of the hot research topics in the field of computer vision. Viewpoint tracking can be applied to image understanding, image compression, image classification, image redirection, information retrieval, etc. [0003] With the development of modern sensing technology and information processing technology, viewpoint tracking technology has also been greatly developed, but it still faces the following problems: [0004] One: In the process of tracking, the viewpoint often faces incom...

AI Technical Summary

Problems solved by technology

[0005] For example, in 2010, Jia Li and other authors published a visual saliency prediction method in the article "Probabilistic Multi-Task Learning for Visual Saliency Estimation in Video" published in the top international journal International Journal of Computer Vision. This method uses multi-task framework to estimate visual saliency, and the multi-task is realized based on low-dimensional visual features and task-related factors. The semantic correlation description between them makes the saliency model unable to restore the global and local information of the image, resulting in the loss of the extracted local or global information of the image;

[0006] Two: Many viewpoint tracking technologies lack the geometric structure between images, resulting in inaccurate tracking

In this article, the adaptive linear regression method is proposed to predict the viewpoint transfer, and the adaptive linear regression method is used to learn a mapping function from high-dimensional features to low-dimensional features of the target feature space, but the method is in the process of predicting viewpoint transfer. Because of the lack of attention to the geometric structure of the salient area of ​​the image, the prediction of the viewpoint can only estimate the fixed viewpoint, and the application range has certain limitations. From an engineering point of view, these methods are not very practical;

[0008] Therefore, so far, there is still no viewpoint tracking method with high tracking accuracy and engineering application.

Images