Robust long-term tracking method based on correlation filtering and target detection

Abstract

The invention provides a robust long-term tracking method based on correlation filtering and target detection, and belongs to the field of target tracking. According to the method, the credibility ofa tracking result is obtained by utilizing a depth feature vector, and a preset credibility threshold value is used for deciding whether a detector is activated or not. When the detector is activated,it will select all targets in the current frame. The most reliable result in all candidate results is obtained by utilizing a multi-stage screening mechanism. Once a new target is obtained, the confidence template is to be updated. According to the method, the interference of an environment object in tracking can be solved by regularly updating the template in tracking.

Description

technical field [0001] The invention belongs to the field of target tracking, in particular to a robust long-term tracking method based on correlation filtering and target detection. Background technique [0002] At present, researchers have solved the ridge regression problem in the frequency domain by using the properties of the circulatory matrix, which greatly speeds up the correlation filtering. However, these filtering algorithms tend to use artificial feature extraction algorithms such as HOG features and gray features. The ability of such feature extraction operators to manually set description images is limited, which leads to fast tracking algorithms, but generally not very good accuracy. The adaptation of deep features solves the problem of insufficient description of image features. The FCNT algorithm analyzes the features of different convolutional layers in detail, and the high-level semantic features are suitable for localization. Low-level detail features ...

AI Technical Summary

Problems solved by technology

If the confidence of the current frame enters this range, the current tracking is considered to be offset

Images