Ternary network-based tight coupling weak supervised learning positioning method and system

Abstract

The invention discloses a tight coupling weak supervised learning positioning method and system based on a ternary network. A positive sample of an image q is obtained by adopting a reordering algorithm based on global features and local features, a negative sample of the image q is obtained by adopting a weak supervision algorithm, and the positive sample and the negative sample serve as a training set of a triple network. The global features and local features are better coupled from two aspects of training sample tuple selection and loss function definition in a learning strategy, mutual promotion of the global features and the local features is realized, and a more effective information representation mode and richer feature details can be learned from a training set; in the training process, the consistency of the two is improved, and the problem of conflict and confusion of the two in results in a visual position identification task is solved, so that the respective advantages are better played in a retrieval framework, and the real-time performance, the high precision and the environment robustness are considered. In addition, the learning strategy also improves the learning efficiency of the model, and greatly reduces the model training time.

Description

technical field [0001] The invention belongs to the field of computer vision and robotics, and in particular relates to a positioning method and system for tightly coupled weakly supervised learning based on a ternary network. Background technique [0002] With the vigorous development of computer vision, retrieval and localization based on deep learning has shown great development potential in the field of robotics. Algorithms are mainly divided into two categories, algorithms based on global features and algorithms based on local features. The algorithm of global features takes a short time to calculate and is invariant to environmental changes, but not to changes in viewing angle; on the contrary, the algorithm of local features takes a long time, is invariant to changes in viewing angles, and has relatively higher accuracy. [0003] Therefore, in order to obtain a high-precision retrieval and positioning scheme that can be processed in real time on the robot, the curren...

AI Technical Summary

Problems solved by technology

[0005] The technical problem to be solved by the present invention is to provide a positioning method and system based on ternary network-based tightly coupled weakly supervised learning, to solve the problem that global features and local features are separated and processed in the two-stage retrieval structure of the current visual position recognition task, Thereby improving the accuracy of search positioning

Images