Adaptive scale tracking method for circular target based on kernel correlated filtering tracking

Abstract

The invention discloses an adaptive scale tracking method for a circular target based on kernel correlated filtering tracking. The method comprises that an image is input, and the resolution of the image is adjusted; a present target tracking area is predicted via a kernel correlated filtering tracking model; an adaptive threshold set is obtained according to the target area, effective edge characteristic points of the target are obtained, edge fitting is carried out, and the position of the tracking area and the resolution of the image are corrected; and the kernel correlated filtering tracking model is updated for next prediction. The problem that the scale of the circular target of kernel correlated filtering tracking cannot be changed adaptively is solved; and the tracking model is more reliable, the online training speed and detection speed of the target model are improved, and tracking is more instantaneous.

Description

technical field [0001] The invention belongs to the field of image target tracking, in particular to a circular target scale adaptive tracking method based on kernel correlation filter tracking. Background technique [0002] Target tracking is the process of performing feature modeling on the detected target area, predicting the position of the target at the next moment, updating the tracking target model, and continuing to predict the position of the target at the next moment. A good object tracking method focuses on improving the robustness, speed and accuracy of the object tracking model. [0003] The target tracking method based on kernel correlation filter has the characteristics of fast tracking speed and high tracking accuracy. However, for some scenarios, there are obvious shortcomings in kernel correlation filter tracking: 1. For scenes with large changes in target scale, the tracking effect is poor. The reason is that the size of the tracking window of nuclear cor...

AI Technical Summary

Problems solved by technology

[0003] The target tracking method based on kernel correlation filter has the characteristics of fast tracking speed and high tracking accuracy. However, for some scenarios, there are obvious shortcomings in kernel correlation filter tracking: 1. For scenes with large target scale changes, the tracking effect The reason is that the size of the tracking window of nuclear correlation filter tracking remains unchanged. If the target scale becomes smaller, a large amount of background interference information will be introduced into the tracking area, polluting the tracking model; if the target scale becomes larger, part of the target information will be lost in the tracking area. , destroying the tracking model, resulting in tracking failure; 2. Unable to judge the tracking failure. For kernel correlation filter tracking, due to the online training modeling method, it is impossible to judge whether the current model is correct, so it is impossible to judge the tracking failure phenomenon caused by the tracking accumulation error ; 3. Cannot eliminate tracking accumulation error, easy to produce tracking drift, and eventually lose track of the target

Images