A Appearance Diversity-Based Generative Adversarial Siamese Network Object Tracking Method

Abstract

The invention discloses a generative confrontation twin network target tracking method based on appearance diversity. For the target video sequence, the initial state of the target is given in the first frame, and the state of the target is estimated in the following video sequence; firstly, for the t-th frame, perform local feature extraction as the search area; where 1<t≤N, N is the total number of frames of the target video sequence; then slide and match the target template in the t-1th frame with the search area extracted in the t-th frame, Obtain the new position of the target; finally, update the target template according to the new position of the target and the current frame, and use the new template as the target template for tracking in the next frame. The invention effectively improves the robustness of the tracker in the face of changes in the appearance of the target, and achieves a better tracking effect.

Description

technical field [0001] The invention belongs to the technical field of computer image processing, and relates to a target tracking method, in particular to a target tracking method based on appearance diversity to generate confrontation twinning networks. Background technique [0002] In recent years, with the development of various image and video hardware technologies, how to extract effective information from images taken by devices is a hot research issue in today's society. Among them, the real-time tracking technology of objects that take moving objects as the research object, because of its problems The complexity and practical significance have attracted the attention of many researchers at home and abroad. As an important branch in the field of computer vision, target tracking technology integrates the knowledge of many traditional disciplines including image processing, pattern recognition, biology, psychology, etc., and in modern times with the vigorous developmen...

AI Technical Summary

Problems solved by technology

The experimental results show that the algorithm proposed by the present invention can effectively solve the problem of the decrease in the robustness of the tracker caused by the diversity of target appearances, and it ranks among the best in the multi-index comparison with several classic algorithms on the OTB100 data set. Excellent tracking results even in complex tracking environments

Images