Background-aware correlation filtering target tracking method

Abstract

The invention provides a background-aware correlation filtering target tracking method. The method is characterized by comprising the steps of firstly, excavating target candidate regions with higherconfidence in a large-range search range, extracting a plurality of cyclic sample groups, and performing filtering detection by using an interference object perception filter; enabling a filter to execute double tasks, selecting a candidate region with the highest confidence degree from all candidate regions, and estimating a preliminary target position on the candidate region through correlationregression; and based on the initial position, realizing fine regression by using a position regression device based on a standard correlation filter to obtain an accurate target position. The background perception learning of the method expands the model perception range, and suppresses the response of an interference region in a targeted manner. Multi-anchor detection effectively expands the target search range, and improves the tracking success rate under the conditions of rapid movement, serious shielding and the like. A progressive target search strategy plays a data alignment effect, sothat the influence of an edge effect is reduced.

Description

technical field [0001] The invention relates to the technical field of image processing and analysis, in particular to a background perception-based correlation filtering target tracking method. Background technique [0002] Object tracking has been an important research topic in the field of image analysis for a long time, and it has very rich applications in the fields of video surveillance, human-computer interaction, navigation and unmanned equipment. In recent years, with the development of visual feature expression and statistical machine learning, the idea of ​​object tracking algorithms has also undergone major changes. It can be roughly divided into two categories: one is generative, this type of algorithm establishes a model based on the criterion of minimizing the target reconstruction error, pays attention to effective modeling of the target appearance, and searches for the target according to the established target appearance model. The highest confidence sampl...

AI Technical Summary

Problems solved by technology

Among them, the most important shortcoming is the edge effect, which greatly limits the performance of the correlation filtering algorithm.

[0004] In addition, the target tracking algorithm also faces many difficulties and problems such as sudden changes in target appearance, sudden changes in motion state, complex video scenes, occlusion, and temporary departure from the field of view.

Images