Target tracking method based on residual dense twin network

Abstract

The invention provides a target tracking method based on a residual dense twin network. The target tracking method comprises the following steps of: firstly, extracting a template image of a to-be-tracked target from a first frame image of a video, inputting the template image into a residual dense network to obtain initial template features, further inputting the extracted features into a globalattention module to obtain template features, and completing tracker initialization; secondly, cutting the t-th frame of image to extract a search region image, and inputting the search region image into the residual dense network to obtain search region features; and finally, inputting the template features and the search region features into a candidate region generation network to obtain a foreground and background classification confidence coefficient and a bounding box regression estimation value, and further acquiring a t-th frame tracking result. By applying the target tracking method and the target tracking device, the problem that an existing target tracking method based on the twin network cannot effectively process background disorder and similar semantic interference is solved,and the problems that an existing target tracking method based on the twin network is low in tracking accuracy and poor in robustness are further solved.

Description

technical field [0001] The invention belongs to the fields of image processing and computer vision, and in particular relates to a target tracking method based on residual dense Siamese network. Background technique [0002] Target tracking refers to automatically and continuously estimating and predicting the position and scale information of the target in subsequent video sequences based on the target to be tracked manually selected in the first frame of the video. Object tracking is a fundamental problem in computer vision, with applications in many fields such as video surveillance, drones, human-machine interfaces, and robot perception. [0003] The target tracking algorithm based on deep learning uses a large amount of labeled data to train the network model offline. Thanks to a large amount of training data, the features extracted by the target tracking algorithm based on deep learning have better expressive power than traditional manual selection features. Tracking ...

AI Technical Summary

Problems solved by technology

[0005] The problem to be solved by the present invention is: in the existing target tracking method based on the twin network, AlexNet is used as the feature extraction network, its feature extraction ability is limited, and it cannot handle the background clutter and similar semantic interference in target tracking well. ; The existing target tracking method based on the twin network only selects the output of the last layer of the network as a feature in the feature selection, which cannot achieve precise positioning of the target; in the tracking process, the deep features of offline training cannot be well adapted to specific targets

Images