Structure sparse tracking method based on significance weighting

Abstract

A structure sparse tracking method based on significance weighting comprises: 1, performing modeling of object appearance, inputting the initial state of an object to obtain an object template and a background template, learning a structure sparse dictionary, performing clustering of local images in an object region, constructing a weight dictionary, and calculating the significance weighting vector, the shielding state vector and the significance weighting structure sparse model of the local image; 2, tracking the object, performing sampling of the object state, obtaining the candidate state of a current frame and a corresponding sampling particle, employing affine transformation to map a candidate particle region into a fixed rectangle, calculating a sample model through the structure sparse dictionary and the weighting vector of the local image, calculating the similarity of the object model and the sample module, estimating the current state of the object according to the maximum posterior probability, and at a temperate update phase, and performing online updating of the object template through a template updating strategy with a shielding detection mechanism to avoid tracking drift so as to better adapt the changing of the object appearance.

Description

technical field [0001] The invention relates to the field of visual tracking, in particular to a structure sparse tracking method based on importance weighting. Background technique [0002] Visual tracking technology is a high-tech subject integrating image processing, pattern recognition, artificial intelligence, automatic control and other fields. It is also one of the key technologies for realizing intelligent robots and intelligent weapons. It is used in military guidance, visual navigation, and security monitoring. , intelligent transportation, video coding, medical diagnosis and meteorological analysis and other fields have broad application prospects and practical significance. The main challenges of visual tracking technology in visual tracking are divided into two categories, one is caused by internal changes in the target, such as target rotation and morphological scale changes, etc.; the other is due to external factors of the target As a result, such as lightin...

AI Technical Summary

Problems solved by technology

The main challenges of visual tracking technology in visual tracking are divided into two categories, one is caused by internal changes in the target, such as target rotation and morphological scale changes, etc.; the other is due to external factors of the target Caused by, such as lighting changes in the scene, occlusion, etc.; and these problems have caused great interference to achieve stable and accurate tracking

Specifically, changes in the shape and scale of the target will lead to huge differences in the performance of the target in different sequence frames; and changes in scene lighting may cause changes in the apparent light and shade of the target, or even severe shadows; in occlusion In the case of the target, it will cause the apparent incompleteness of the target or introduce noise to the target area; the rapid change of the target in the scene will cause motion blur, so that the target cannot be easily distinguished; and the background clutter in the scene will also lead to It is difficult to distinguish the target from the background information, which affects the tracking effect

Images