Visual tracking method combining classification and domain adaptation

Abstract

The invention discloses a visual tracking method combining classification and domain adaptation, on one hand, the advantages of a classification composition method and a transfer learning domain adaptation method are combined, and a feature space which can separate a foreground from a background and can be shared by recent acquired source domain data and current to-be-classified target domain datais sought; Another aspect, a deep learning idea is adopted; the method has a prospect in a target domain Xt; in the background judgment and feature space mapping matrix V updating process, the feature space mapping matrix V is updated; layer-by-layer iteration is carried out, the combination of classification composition and domain adaptive learning is completed according to the result of the previous iteration in each iteration, the output result V of the current layer is used as the input of the next iteration, and a tracking result is obtained through layer-by-layer iterative learning, sothat the accuracy of the tracking result is higher.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and more specifically relates to a visual tracking method combining classification and domain adaptation. Background technique [0002] With the development of artificial intelligence and machine learning technology, visual tracking has always been a research hotspot in the field of computer vision technology. Visual tracking is the process of position detection, feature extraction and recognition, and tracking of targets in videos. It can be applied in many fields such as video surveillance, human-computer interaction, augmented reality, intelligent transportation, and video compression. Since the tracking target often has characteristics such as displacement, deformation, and rotation, and the complexity of the surrounding environment, such as changes in light, obstacles, and changes in reference objects, visual tracking has always been a challenging research topic. [0003] The classi...

AI Technical Summary

Problems solved by technology

This method is simple and clear, but it usually needs to know the data feature distribution of the foreground and background in advance, so supervised or semi-supervised learning methods are often used, while the training sample data in the real environment is often not known in advance, making supervised or semi-supervised Supervised learning is hindered

Images