A tracking algorithm based on high confidence update supplementary learning

Abstract

The invention discloses a tracking algorithm based on high confidence update supplementary learning, which belongs to the field of image processing. At first, the correlation filter response value isobtained according to the standard correlation filter classifier, and the peak-to-side ratio of the correlation filter response is calculated as a response confidence level; if the confidence level isgreater than an average threshold, the current frame continues to update the correlation filter, and if the confidence level is less than an average threshold, the update of the filter is stopped; then, the number of frames that are not updated continuously is calculated, and if there are 10 consecutive frames that are not updated, the frames are updated forcibly; finally, the total response is obtained by fusing the response of the color supplementary learner, and the position of the maximum value in the response is the tracking result. The invention remarkably improves the robustness of thetracking algorithm, can effectively distinguish the target from the background, further improves the precision of the tracker, and remarkably improves the robustness and accuracy of the tracker underthe condition that the illumination changes drastically and the target scene is complex. The algorithm is free of the influence of the change of the tracking target environment, and can effectively and accurately track the target object.

Description

technical field [0001] The invention relates to a tracking algorithm based on high-confidence update supplementary learning, especially a video tracking method, which belongs to the field of image processing. Background technique [0002] Object tracking is one of the core problems in the field of computer vision, and it has a wide range of applications in human-computer interaction, video surveillance, augmented reality and other fields. Although a lot of progress has been made in this field in the past few decades, tracking arbitrary objects is still a very challenging task due to disturbances such as illumination changes, geometric deformations, partial occlusions, background clutter, fast motion, etc. [0003] In recent years, many target tracking algorithms based on correlation filtering have emerged, which can perform better single target tracking in video, among which the real-time color supplementary learning target tracking algorithm is representative. However, the...

AI Technical Summary

Problems solved by technology

[0004] The technical problem to be solved in the present invention is to solve the problem that the existing target tracking algorithm does not consider the reliability of the current frame result when updating the filter, which leads to tracking failure. Based on the supplementary learning tracking (Staple), a high-confidence detection step is added to make the tracking algorithm more robust and accurate

Images