Twin network visual tracking method based on memory unit

Abstract

The invention discloses a twin network visual tracking method based on a memory unit, which belongs to the technical field of target tracking, and comprises the following steps of 1, establishing a tracking model, 2, obtaining initial target template features, 3, obtaining the corresponding position of the tracking target in the current frame of the video, 4, according to the position of the tracking target in the current frame found in the step 3, cutting out an area where the target is located as a target template, and inputting the target template into a target template branch of a trackingmodel to obtain a new target template feature, and reading the next frame of the video as the current frame, and turning to the step 3 to carry out the next round of iteration until all frames in thevideo are read and iteration is finished. According to the method, the problems of shielding, background mixing, violent target form change and the like in the visual tracking process can be effectively solved, and the tracking robustness of the tracking model facing a complex environment is improved.

Description

technical field [0001] The invention belongs to the technical field of target tracking, and in particular relates to a twin network visual tracking method based on a memory unit. Background technique [0002] Object tracking technology plays a pivotal role in computer vision technology and is widely used in fields such as smart transportation, security, sports, medical care, robot navigation and human-computer interaction, and has great commercial value. The task of visual tracking is to select an area of ​​interest in an image sequence as a tracking target, and obtain accurate target position, specific shape, and target trajectory information in the next several consecutive image frames. From the perspective of technological development, the research on visual tracking technology can be divided into three stages. In the first stage, the classic tracking methods represented by Kalman filter, mean filter, particle filter and optical flow method; in the second stage, The dete...

AI Technical Summary

Problems solved by technology

[0004] However, in the traditional Siamese network tracking model, only the initial frame is used as the target template, and the network parameters and target template of the model are no longer updated after the offline training is completed.

If the network parameters are not updated, it means that the tracking performance of the model will suffer a huge decline when encountering unseen scenes or targets. If the target template is not updated, the target in the video sequence will undergo drastic appearance changes or be severely occluded. Generate tracking drift problems, which will lead to a decline in model robustness and tracking accuracy

Images