Flexible updating strategy for target tracking

Abstract

The invention discloses a target tracking method of an elastic updating strategy, which takes a SAMF algorithm as a reference and combines an APCE threshold value and an APCE gradient threshold valueon the basis of an average peak correlation energy (APCE) to judge the reliability of the tracking result so as to determine the model updating speed. The APCE threshold is enhanced inversely and theAPCE gradient threshold is enhanced positively, and the APCE and APCE gradients are updated when the APCE and APCE gradients are higher than the respective thresholds, otherwise the update rate is adjusted according to the APCE and its gradient variation. The gradient change of APCE is used to fully reflect the change of the disturbed process of the target. Based on this, different update rate, i.e. Elastic update strategy, is adopted to achieve better processing ability for the fast movement and partial occlusion of the target.

Description

technical field [0001] The invention belongs to the technical field of video target tracking, and relates to a target tracking method, in particular to a target tracking method with an elastic update strategy. Background technique [0002] Object tracking is an attractive and rapidly developing field in computer vision, which involves many challenging research hotspots and often occurs in conjunction with other computer vision problems, such as human-computer interaction, video surveillance, augmented reality, autonomous driving, and mobile robotics. and many more. In recent years, Correlation Filters (CF for short) have been introduced into the framework of object tracking, and have achieved remarkable results in both accuracy and speed. In 2010, Bolme et al. proposed a new correlation filter, MOSSE (Minimum Output Sum of Squared Error), and applied CF to the tracking algorithm for the first time. Object tracking is formulated as a correlation filtering problem equivalent...

AI Technical Summary

Problems solved by technology

The advantage of this method is intuitive and simple, but when the interference target appears, there will be multiple peaks in the response graph, and the response value of the interference target may be greater than the real target. At this time, if the target is updated, the model will be affected by the interference target. drift

[0006] (2) Peak sidelobe ratio (PSR) judgment method: This method uses the ratio of the peak value to the sidelobe as a measure. When the interference target appears, although the maximum peak value is large, the peak sidelobe rate is low

However, the common disadvantage of both is that they can only judge the result when the final target is disturbed, but cannot judge the process change of the target being polluted.

When the values ​​of the two suddenly become smaller, it means that the target has been severely disturbed at this time, and the update rate should be adjusted appropriately. When the smaller value is not enough to reach below the threshold, it is not very good to use these two methods to judge at this time. Reasonable

Images